[{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Spot and preemptible instances offer substantial per-hour cost reductions for cloud compute but introduce job interruption risk that can increase total cost when checkpointing is not implemented. We built a cost model for WGS alignment and variant-calling pipelines that incorporates hourly spot price distributions sampled over 90 days from two cloud providers, mean time to interruption distributions per instance family, and checkpoint overhead for BWA-MEM2 and GATK stages. Simulations over 500 synthetic cohort scenarios (50–1000 samples, 10×–30× coverage) showed that naive use of spot instances without checkpointing increased effective cost by 18–34 % relative to on-demand pricing in high-volatility markets. A bidding and checkpoint strategy derived from the model reduced pipeline cost by 41 % on average relative to on-demand in stable price periods. We validated the model against actual billing records from a 200-sample pilot run. The model is implemented as a Python library with a command-line interface.","excerptSha256":"059e4102dc96bc777641b69d3520d9624aff75caad280df8a5e5b847a7776fe2","id":"ev-17eb6d3e27a3376b2f2e3e4d515dba94c575628055a59c5a949a8afe31768b2f","recordSha256":"3df5f68c186961d19b3857b894012142e82c6eba783f020d37d87976012769f5","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1615},"recordId":"SYN-PUB-01615"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0123","excerptSha256":"aeec523febcad55bf7c251a627cdfee8c7be32d42415a4ea94e5b8035bb501b8","id":"ev-127c213f3b14f099f76885e8f4d49b21e3dc6379799d497ca809e517dd364441","recordSha256":"3df5f68c186961d19b3857b894012142e82c6eba783f020d37d87976012769f5","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1615},"recordId":"SYN-PUB-01615"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"read alignment\",\"variant calling\",\"pipeline parallelisation\",\"cloud cost optimisation\",\"clinical genomics\",\"WGS/WES processing\",\"HPC scheduling\",\"workflow management\",\"GATK\",\"genomic data compression\"]","excerptSha256":"54640a5b9786bd5ba3dd54754de5309dcd8b61aebc174d6a69f25e5d5ad1cce3","id":"ev-625978e55e235393f924c2e318d2719d442f8dbb4c303d42744eac77a938f3e5","recordSha256":"2dc4dd53e011f0bad53f9f5c3506ffb2dd2445f6589387911d064cfe61d74048","source":{"derivedRecordId":false,"fieldPointer":"/keywords","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/122"},"recordId":"R0123"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0123","excerptSha256":"aeec523febcad55bf7c251a627cdfee8c7be32d42415a4ea94e5b8035bb501b8","id":"ev-cdb8e426cbcb7aa46dc70b797a21f69cced88b7fecf973b196093ce40396c956","recordSha256":"2dc4dd53e011f0bad53f9f5c3506ffb2dd2445f6589387911d064cfe61d74048","source":{"derivedRecordId":false,"fieldPointer":"/id","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/122"},"recordId":"R0123"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"GENOMFLO addresses the high computational cost and long turnaround time of clinical WGS and WES pipelines at research hospitals in Serbia and the region. The project develops and validates a cloud-cost-aware orchestration layer that dynamically selects between on-premises HPC and spot-instance cloud execution depending on queue state, sample priority, and real-time cloud pricing. It also provides standardised containerised workflow components for alignment and variant calling that can be deployed by laboratories with limited bioinformatics staff. Pilot deployments are planned at two regional clinical partners. The project includes training activities for clinical bioinformaticians and releases all software under an open-source licence.","excerptSha256":"19f93b2e210ec8a4ac65a0a5405424b85dfd8d154a608013a0270b07504e0adc","id":"ev-c7e8de2c805b759e633e2f6328735e9f087252e1884af5370d37272fa10162f5","recordSha256":"8cc9b8a49c45976b045457ddba9ecc695d066a8ad2b876530939168bf00fbf28","source":{"derivedRecordId":false,"fieldPointer":"/summary","file":"data/projects.json","locator":{"format":"json","jsonPointer":"/projects/253"},"recordId":"SYN-PRJ-0254"},"sourceFileSha256":"847cb13091f91ed9fe57f6aea0088cd1e66ed0263c86a5090f169a2b30ab3a1a"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0123","excerptSha256":"aeec523febcad55bf7c251a627cdfee8c7be32d42415a4ea94e5b8035bb501b8","id":"ev-5f1f974c5f0df9b942354f61a09d05cf882becdcd97f0825ff9f33df178b195e","recordSha256":"8cc9b8a49c45976b045457ddba9ecc695d066a8ad2b876530939168bf00fbf28","source":{"derivedRecordId":false,"fieldPointer":"/participants/0/researcher_id","file":"data/projects.json","locator":{"format":"json","jsonPointer":"/projects/253"},"recordId":"SYN-PRJ-0254"},"sourceFileSha256":"847cb13091f91ed9fe57f6aea0088cd1e66ed0263c86a5090f169a2b30ab3a1a"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Alignment accuracy at structural variant breakpoints affects downstream SV calling quality, yet most aligner benchmarks focus on SNPs and small indels. We constructed a synthetic benchmark dataset of 2 200 breakpoint-spanning read pairs spanning 110 curated SV loci (deletions, tandem duplications, and inversions ranging from 50 bp to 50 kb) derived from a validated clinical variant database. Four short-read aligners were evaluated: two BWT-based and two graph-based tools. We measured mapping quality distribution, soft-clip rate within 200 bp of breakpoints, and downstream SV call concordance using two SV callers. Graph-based aligners achieved 12–18 percentage point higher recall at inversion breakpoints compared to BWT-based aligners, but produced 2.3× more multimapping reads genome-wide, increasing false-positive SV calls by 9 %. Runtime and memory differences between aligner classes were substantial: graph-based alignment required 3.1× more peak RAM and 2.4× more wall time per sample. Results are tabulated per SV class and size bin.","excerptSha256":"dd065d73e560a2fc2dd3e91767f04b178df846ffb197e2abd1c8befd4ff88f4f","id":"ev-b883d6b162209ccffcea999a4f248970d03cc2988f8946f05ca2ee73c3871e94","recordSha256":"2a0a90f8a561bfc983e66ce5a8448e07932e205d317824c5ef10fc06b680d825","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1617},"recordId":"SYN-PUB-01617"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0123","excerptSha256":"aeec523febcad55bf7c251a627cdfee8c7be32d42415a4ea94e5b8035bb501b8","id":"ev-05251d37d0b4d8c9232e0be18e430a795588823464140787a27531dfd48aff05","recordSha256":"2a0a90f8a561bfc983e66ce5a8448e07932e205d317824c5ef10fc06b680d825","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1617},"recordId":"SYN-PUB-01617"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"GATK HaplotypeCaller remains a bottleneck in clinical WES pipelines because its local de novo assembly step is largely single-threaded at the per-sample level. We systematically evaluated three strategies for throughput improvement on a cohort of 412 anonymised WES samples (mean coverage 85×, 38–62 Mb target): genomic-interval scatter–gather with 100 Mb shards, sample-level parallelism across nodes, and a combination of both. Job orchestration used Snakemake with a SLURM backend. The scatter–gather approach alone reduced mean per-sample runtime from 4.1 hours to 41 minutes on 32 cores. Combining it with sample-level parallelism across 8 nodes processed the full 412-sample cohort in 19 hours wall time compared to 70 hours in the sequential baseline. We measured concordance against the baseline using precision and recall on GIAB benchmark variants: SNP precision 99.91 %, recall 99.87 %; indel precision 99.34 %, recall 99.21 %. Configuration files and Snakemake rules are provided as supplementary material.","excerptSha256":"a7cdd1fb7d19ff67e74a23ef4a815a9694038ae78119f6d16924edb71e4d2eb7","id":"ev-386ebffa1089ce50395b10e140c2bf00d9432f00b34aeb4c55f9d96b8cbee020","recordSha256":"c1e41d568783dbf659543740d86e915b8e2205e315ffb01d2850361dd854ae49","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1614},"recordId":"SYN-PUB-01614"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0123","excerptSha256":"aeec523febcad55bf7c251a627cdfee8c7be32d42415a4ea94e5b8035bb501b8","id":"ev-6d8d59a890ebaab690cdd83583375e0c20744530a7560717c74f0deddebb53fc","recordSha256":"c1e41d568783dbf659543740d86e915b8e2205e315ffb01d2850361dd854ae49","source":{"derivedRecordId":false,"fieldPointer":"/authors/3/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1614},"recordId":"SYN-PUB-01614"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"cloud cost\",\"spot instances\",\"WGS pipeline\",\"checkpointing\",\"cost modelling\"]","excerptSha256":"ae33d82b1493555bfb8ddfdb993c3fd223e66564ffa5b2b524621b16eb6d5aa4","id":"ev-e2936af7872e400217df9e2d304f79c3990e5532e132eda75307dfdb3afc7aa8","recordSha256":"3df5f68c186961d19b3857b894012142e82c6eba783f020d37d87976012769f5","source":{"derivedRecordId":false,"fieldPointer":"/keywords","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1615},"recordId":"SYN-PUB-01615"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Boris Rakić received his B.Sc. in Computer Science from the University of Novi Sad in 2011 and his M.Sc. in Bioinformatics from the same institution in 2014. He completed his Ph.D. at the University of Novi Sad in 2019, with a thesis on parallelising short-read alignment algorithms for shared-memory architectures. Since 2020 he has been a Research Associate in the Data Science and Bioinformatics Group at BioSense Institute, where the group develops scalable data pipelines for precision agriculture and clinical genomics. He contributes to teaching a graduate-level course on biological data analysis. He works primarily with the group's on-premises HPC cluster and has hands-on experience with AWS and Google Cloud genomics environments.","excerptSha256":"17f0a284b5eb1a4012abcca7e2a70d30f9e84510a87fda2fbd08ba142c6f419a","id":"ev-10ea081c1d4ce5d21da7b366caba15ed185c103fa10706ccbc59178341f15d0c","recordSha256":"2dc4dd53e011f0bad53f9f5c3506ffb2dd2445f6589387911d064cfe61d74048","source":{"derivedRecordId":false,"fieldPointer":"/bio","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/122"},"recordId":"R0123"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"Bioinformatics\",\"High-Performance Computing\",\"Genomics\",\"Cloud Computing\"]","excerptSha256":"d8ceb9e8170d7ab2623b002c097289a45321af63a163ba780c74149b9df6ecb0","id":"ev-54343b077a33e039b3a4a0b795a33ae0741442ddfdf57958b684a6accb6f84b8","recordSha256":"2dc4dd53e011f0bad53f9f5c3506ffb2dd2445f6589387911d064cfe61d74048","source":{"derivedRecordId":false,"fieldPointer":"/fields","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/122"},"recordId":"R0123"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"structural variants\",\"read alignment\",\"benchmark\",\"graph genome\",\"SV calling\"]","excerptSha256":"cd1fa57da601bd068bba18d14817dc49c16dac9d8f52fdd2c12c3f7f75765573","id":"ev-bd07b1ac1f6c276eb12faa6ea2f8fa5c8a2e589a742312b59628a4c0d8081cb3","recordSha256":"2a0a90f8a561bfc983e66ce5a8448e07932e205d317824c5ef10fc06b680d825","source":{"derivedRecordId":false,"fieldPointer":"/keywords","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1617},"recordId":"SYN-PUB-01617"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"University HPC clusters frequently host both long-running genomic assembly jobs and short, interactive variant-calling tasks submitted by clinical collaborators. We ran a six-month measurement study on a 640-core cluster at the University of Novi Sad, logging 18 400 job submissions across three user groups. We then replayed the trace under five SLURM scheduling configurations—backfill-only, priority preemption, gang scheduling, fair-share with decay factor 0.5, and a hybrid policy—using a discrete-event simulator we built in Python. The hybrid policy combining fair-share weighting with selective preemption of low-priority assembly jobs reduced median wait time for clinical variant-calling tasks from 47 minutes to 11 minutes, at the cost of a 3 % increase in total cluster makespan over the trace period. We describe configuration parameters, discuss trade-offs for clusters with heterogeneous node types, and release the simulator and anonymised job trace under an open licence.","excerptSha256":"b857d3ea8250f8178a6f79874caa0bb9c4edd1e9114e1a8bda1edd75a29697a4","id":"ev-abeb4cbb8f52a9f1e760516ad6ae95e74edbda48e9c5621066d078d132b74dae","recordSha256":"abc3c0dca670013f194dd2bdbf02517156259fcd617c07553c328803bf1c7f77","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1613},"recordId":"SYN-PUB-01613"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0123","excerptSha256":"aeec523febcad55bf7c251a627cdfee8c7be32d42415a4ea94e5b8035bb501b8","id":"ev-3700be8240f7a41da844bba1ce6c8d1be13aa443157d558fb1becb52d452005f","recordSha256":"abc3c0dca670013f194dd2bdbf02517156259fcd617c07553c328803bf1c7f77","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1613},"recordId":"SYN-PUB-01613"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Cache-Oblivious Data Layouts for Short-Read Alignment on Shared-Memory Multicore Systems","excerptSha256":"a22f6acbe8fae1515de9f9d081d7e67124f9e5403f4e5c7141bb98ebac447333","id":"ev-756362a0e60229f6e2fa04f25c74f98b4647f4985d053fae0527384acb43b07d","recordSha256":"2f71acb25030121f43566b32898a3092fd73b441437b114b4e6fbf1c3febebdf","source":{"derivedRecordId":false,"fieldPointer":"/title","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1612},"recordId":"SYN-PUB-01612"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0123","excerptSha256":"aeec523febcad55bf7c251a627cdfee8c7be32d42415a4ea94e5b8035bb501b8","id":"ev-c00aae10fa73cbe3f07a4472e44675ccd7548f6fe08140a36a0cd3e2b9843a0c","recordSha256":"2f71acb25030121f43566b32898a3092fd73b441437b114b4e6fbf1c3febebdf","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1612},"recordId":"SYN-PUB-01612"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Soft-clipped bases in BAM files are stored as raw sequence and contribute disproportionately to file size in samples with adapter contamination or structural-variant-rich regions. We characterised soft-clip content in 60 WGS BAM files (mean 30× coverage, 150 bp reads) and found that soft-clipped bases account for 9–23 % of stored bases but represent only 2–6 % of aligned bases. We designed a reference-relative encoding scheme that replaces soft-clipped sequences with a compact difference representation against a local reference window, embedded within a standard BAM extension tag so that existing downstream tools read the files without modification. Across the 60-file benchmark, the scheme reduced total BAM size by 7.4 % on average with a maximum of 14.1 % in the most clip-heavy sample. Encoding and decoding add 4 % and 2 % respectively to BAM I/O time. The codec is implemented as a standalone C utility compatible with htslib.","excerptSha256":"492606b2023e27a6620a79bc4c4fbda9df383ce131db1fa9a2b495b16a527eb2","id":"ev-71a84ab4ae6f3a1ed0574e4f182a2d0670713493ac5fe438a092f8be7525847f","recordSha256":"745e855d2fcde6da12dc8788a0135f46a872a443823f9184073dc60ef43118fd","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1616},"recordId":"SYN-PUB-01616"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0123","excerptSha256":"aeec523febcad55bf7c251a627cdfee8c7be32d42415a4ea94e5b8035bb501b8","id":"ev-a81568da8d0c511fd5c641fd2d8ede0450c30b66685db0851ecf199e46869bae","recordSha256":"745e855d2fcde6da12dc8788a0135f46a872a443823f9184073dc60ef43118fd","source":{"derivedRecordId":false,"fieldPointer":"/authors/3/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1616},"recordId":"SYN-PUB-01616"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"He is particularly interested in partnerships with companies that generate or process genomic data at scale and face concrete cost or throughput constraints; he is less interested in purely theoretical or database-curation collaborations.","excerptSha256":"991a9d641a7c1ac060830c2a6d171c6e1487362ea429d4571477066c261d4d49","id":"ev-51a912e3c4cc6026d4b2e03aa571708f3b6c2b9c32a275d354e47e28c4bfe9db","recordSha256":"2dc4dd53e011f0bad53f9f5c3506ffb2dd2445f6589387911d064cfe61d74048","source":{"derivedRecordId":false,"fieldPointer":"/notes","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/122"},"recordId":"R0123"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"joint projects\",\"contract research\",\"consulting\",\"student internships\"]","excerptSha256":"1f80e0a1a3b6707c62dd48394792f7c19211bdf1ecbbf54f4ed5873d76bf75d7","id":"ev-730c8d9381c57c0bd5c7a3f5044e4621d32878f8022ab5fcef6c0158c7fd7baa","recordSha256":"2dc4dd53e011f0bad53f9f5c3506ffb2dd2445f6589387911d064cfe61d74048","source":{"derivedRecordId":false,"fieldPointer":"/open_to","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/122"},"recordId":"R0123"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Boris Rakić completed a three-month consulting engagement in 2023 with a Novi Sad-based health-tech startup to audit and optimise their cloud-based WES processing pipeline, reducing their per-sample cloud spend by approximately 35 %.","excerptSha256":"c850ce2fbcc7c9eca9bec03a6f87310543e0b6ee237bebdb93073790c07ca921","id":"ev-226e5222e55fdfeb9a9aa56e0f49f3737b4529536d32a22e8158d3bfed47f3f4","recordSha256":"2dc4dd53e011f0bad53f9f5c3506ffb2dd2445f6589387911d064cfe61d74048","source":{"derivedRecordId":false,"fieldPointer":"/industry_experience","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/122"},"recordId":"R0123"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0781","excerptSha256":"c8229d43fe347190149565eda439f3e6fe52c733342d687d5b1ea0ea059ed360","id":"ev-67c4f22ec61fb3e0b3f6522d557eff7a681686a3d7369124560d07320124fe14","recordSha256":"c1e41d568783dbf659543740d86e915b8e2205e315ffb01d2850361dd854ae49","source":{"derivedRecordId":false,"fieldPointer":"/authors/1/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1614},"recordId":"SYN-PUB-01614"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Reproducibility in single-cell RNA sequencing analyses depends not only on algorithmic choices but also on the computational environment, including workflow manager, container version, and execution platform. This study ran an identical scRNA-seq clustering pipeline—from raw count matrix to annotated UMAP—using Snakemake, Nextflow, WDL, CWL, and a custom Makefile-based approach across three computational environments: a local HPC cluster, a cloud instance, and a local workstation. The pipeline was applied to two datasets of 8,000 and 22,000 cells. Cell-type cluster assignments were identical across workflow managers within the same environment for deterministic steps but diverged by up to 4.1% of cells in stochastic steps when floating-point rounding behaviour differed between platforms. Conda environment pinning and container-based execution reduced inter-platform divergence to under 0.3% of cells. A minimum specification for reproducible scRNA-seq pipeline reporting was proposed, covering software versions, container hashes, and random seed logging.","excerptSha256":"f42644e5e65779ca17f24be3e8faddfae0fc643368348250dd7a37ce7e9ac038","id":"ev-7d79567f956cf84b1f6f06dd57b45f742505cec099f1fae4eaba4815c14e2501","recordSha256":"a2fa8e9cb3cd8b8cfb9a91186329cd5b50cca65224552b4ca3a561c127a72ecf","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8565},"recordId":"SYN-PUB-08565"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0781","excerptSha256":"c8229d43fe347190149565eda439f3e6fe52c733342d687d5b1ea0ea059ed360","id":"ev-44e0b811482cfbe78d75fea86603ec943ca5b77e2bf998007da89e56f36cd09e","recordSha256":"a2fa8e9cb3cd8b8cfb9a91186329cd5b50cca65224552b4ca3a561c127a72ecf","source":{"derivedRecordId":false,"fieldPointer":"/authors/1/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8565},"recordId":"SYN-PUB-08565"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"GATK\",\"HaplotypeCaller\",\"WES\",\"scatter-gather\",\"Snakemake\",\"variant calling\"]","excerptSha256":"1084c899dfb03816bf6310975ddaa7c5d5b4de95fa03187b637467d42e4c391d","id":"ev-a294788f2e4d82b88a76bbc7a1c7fb1f22fb3a91d99585b1c00ad33c9ad96346","recordSha256":"c1e41d568783dbf659543740d86e915b8e2205e315ffb01d2850361dd854ae49","source":{"derivedRecordId":false,"fieldPointer":"/keywords","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1614},"recordId":"SYN-PUB-01614"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"reproducibility\",\"workflow managers\",\"Snakemake\",\"Nextflow\",\"pipeline benchmarking\"]","excerptSha256":"66ee3ece6d3c4dec6c70668b782cb706c9bc40df941a3e32d98b0e69cccc532b","id":"ev-55c6dc2c253c2a841318c4f42f67998fb9421d170f3e0507fbbb47a5a8299df5","recordSha256":"a2fa8e9cb3cd8b8cfb9a91186329cd5b50cca65224552b4ca3a561c127a72ecf","source":{"derivedRecordId":false,"fieldPointer":"/keywords","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8565},"recordId":"SYN-PUB-08565"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"Bioinformatics\",\"Machine Learning and Probabilistic Modelling\",\"Plant Biology and Genomics\",\"Data Science\"]","excerptSha256":"9c0bcf6a5f2ba3d8fd2e4a8eb8e7d6906d08b00aa04e718c3f2eb1324e7a0e8d","id":"ev-ebe37a1fc9b0495d171f86fa249399189f2adb5c5bad6bf1c8801dee11bcc856","recordSha256":"39166eb2bbbbc7ff16474c84a934f4870dba84e84304deb21b9446541ecf1d04","source":{"derivedRecordId":false,"fieldPointer":"/fields","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/780"},"recordId":"R0781"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0781","excerptSha256":"c8229d43fe347190149565eda439f3e6fe52c733342d687d5b1ea0ea059ed360","id":"ev-d4bf8d8db5e5cd9da0fc653c176145c8e8e8dd8d69eb9ee899d29d2b7dab6511","recordSha256":"39166eb2bbbbc7ff16474c84a934f4870dba84e84304deb21b9446541ecf1d04","source":{"derivedRecordId":false,"fieldPointer":"/id","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/780"},"recordId":"R0781"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"A Lightweight Pipeline for Extracting Drug–Disease Relations from Serbian Clinical Case Reports","excerptSha256":"4cd11bf59f594c4264d9bde4859ef027bce8248e292e151137fb30af37fb40fc","id":"ev-523e814b02a44489812bd9cd0e66eb1738f3cf2d27b6f58642ba065fab4e4c60","recordSha256":"af6a362ad1e198effcbde9b0aa0ec5a0a2a1cf0e7eac368d36c028fab2bdea7c","source":{"derivedRecordId":false,"fieldPointer":"/title","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6256},"recordId":"SYN-PUB-06256"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0781","excerptSha256":"c8229d43fe347190149565eda439f3e6fe52c733342d687d5b1ea0ea059ed360","id":"ev-b0bbeb13eab50873303cb33aa39b004022ac122468bcc1771637f61ca5828ac1","recordSha256":"af6a362ad1e198effcbde9b0aa0ec5a0a2a1cf0e7eac368d36c028fab2bdea7c","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6256},"recordId":"SYN-PUB-06256"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Relation extraction between biomedical entities in low-resource languages is typically approached by adapting English-trained models, but direct adaptation is complicated by the absence of parallel corpora and the syntactic differences of inflected languages. This preprint describes a lightweight relation extraction pipeline applied to a set of 620 Serbian clinical case reports collected from three open-access Serbian medical journals between 2015 and 2025. The pipeline uses the disease and gene named entity recogniser developed in prior work, extended here with a drug entity module trained on a manually annotated set of 300 sentences. Candidate drug–disease pairs are classified using a support vector machine over typed dependency features extracted from a Serbian dependency parser. The approach avoids large model fine-tuning and can run on CPU hardware. On a held-out set of 80 case reports, the pipeline identifies drug–disease treatment relations with a precision of 0.72 and recall of 0.61. Error analysis indicates that the main precision losses arise from non-treatment drug mentions such as contraindications mentioned in passing. The code and evaluation data are shared as a preprint to invite community feedback before journal submission.","excerptSha256":"f090ea0e91b69460bcefe28947290d483fe48f8307bff128d27528de06cbe36b","id":"ev-2d84d5470148fa393c2fc7eed8ed1491409994153465088a3738b97cc84e93e4","recordSha256":"af6a362ad1e198effcbde9b0aa0ec5a0a2a1cf0e7eac368d36c028fab2bdea7c","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6256},"recordId":"SYN-PUB-06256"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Parallelisation Strategies for GATK HaplotypeCaller in Clinical Whole-Exome Sequencing Cohorts","excerptSha256":"f0247c424a39a34dbe403ac11d7e8225451428509c3f2865b91e61f4d1189677","id":"ev-74cc130c10347c2c91319c6e5bda5e13efec09b2037f3fadd620b90e04dd5e39","recordSha256":"c1e41d568783dbf659543740d86e915b8e2205e315ffb01d2850361dd854ae49","source":{"derivedRecordId":false,"fieldPointer":"/title","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1614},"recordId":"SYN-PUB-01614"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Ivona Marković completed her BSc in Mathematics and Computer Science at the University of Novi Sad in 2013, followed by an MSc in Bioinformatics at the same institution in 2015. She defended her PhD at the Faculty of Sciences, University of Novi Sad, in 2020, with a thesis on structure learning of Bayesian networks under limited sample conditions applied to plant genomics. Since 2021 she has held a Research Associate position in the Data Science and Bioinformatics Group at BioSense Institute, where the group develops computational methods for high-throughput agricultural and environmental data. She occasionally co-supervises master students and delivers guest lectures in probabilistic modelling. Her primary tools include high-performance computing clusters at BioSense, R and Python environments, and she regularly accesses RNA-seq datasets from greenhouse stress-experiment platforms maintained at the institute.","excerptSha256":"402b4cf5a04d1f1b0af5d690d07213f8b47376870544505faad5d063a6909a25","id":"ev-f96994b5cb2813c5e77424a954be06031b1b7eda8efdec6d14696726710941f9","recordSha256":"39166eb2bbbbc7ff16474c84a934f4870dba84e84304deb21b9446541ecf1d04","source":{"derivedRecordId":false,"fieldPointer":"/bio","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/780"},"recordId":"R0781"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Inferring gene regulatory networks from observational transcriptomic data is complicated when sample sizes are small relative to the number of candidate variables. This study evaluated four score-based structure learning algorithms—BDe, BGe, BIC-penalised hill-climbing, and a tabu-search variant—on RNA-seq data from 18 Arabidopsis thaliana samples subjected to 4°C cold stress over 24 hours. We assessed edge recovery against a manually curated reference subnetwork of 42 known regulatory interactions. The BGe criterion with a conservative prior on graph complexity recovered 61% of true edges at a false discovery rate of 0.18, outperforming BIC-penalised search (52% recovery, FDR 0.24). We also examined sensitivity to prior specification by varying hyperparameters across a grid of 120 settings. Results indicate that informative priors derived from published protein–protein interaction data improve skeleton accuracy by approximately 9 percentage points in this low-n regime. Data and scoring code are deposited in a public repository.","excerptSha256":"5c6dcbb7c801d123c7753b18ac03e6351d1879a37eebc737bd84d1198fb9c086","id":"ev-d1780c832888e621d075595bd41ac4ad54aae7a3acd40fb05a80ec915d4891b8","recordSha256":"4c537c348462aa716583c630fa18784438949d0f334da1adf47f6e745135bf23","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8558},"recordId":"SYN-PUB-08558"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0781","excerptSha256":"c8229d43fe347190149565eda439f3e6fe52c733342d687d5b1ea0ea059ed360","id":"ev-220735b78131eec173b50ba3779f371ae0bd57f69a9920b80a9a999eb9c45878","recordSha256":"4c537c348462aa716583c630fa18784438949d0f334da1adf47f6e745135bf23","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8558},"recordId":"SYN-PUB-08558"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"A large number of open-source packages implement Bayesian network structure learning, yet systematic benchmarks on biological transcriptomic data in the low-sample regime are rare. We evaluated eight packages—spanning R and Python ecosystems—on 90 real RNA-seq datasets collected from plant stress experiments available in public repositories, filtered to those with 15–48 samples. Packages were assessed on runtime, reproducibility of results across operating systems, flexibility of prior specification, and accuracy measured by SHD against gold-standard subnetworks. Bnlearn (R) and pgmpy (Python) offered the broadest algorithmic coverage; bnlearn recovered edges with the highest F1 (0.41 mean across datasets) while pgmpy was 2.3× faster on datasets exceeding 200 variables. Three packages produced non-reproducible results due to undocumented random seeds. We provide a reproducible benchmarking harness and a decision guide aimed at plant biologists with limited computational experience.","excerptSha256":"12caceebcff531161177000a1a5e58aece80a07288c19f9958a91a84b2060d51","id":"ev-2ae2876a11d2b609d369ac5dd742ea134c9ccaa9093ebf71b248e3899aed2a39","recordSha256":"e60eece84f5b7d1e4196e23b1961475f7276f49579e13454de3b141399c96e2a","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":2159},"recordId":"SYN-PUB-02159"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0781","excerptSha256":"c8229d43fe347190149565eda439f3e6fe52c733342d687d5b1ea0ea059ed360","id":"ev-4a6a281adc4592f193491675aa0791a77bb4eef7527c8dce6835d1bfab6dcb05","recordSha256":"e60eece84f5b7d1e4196e23b1961475f7276f49579e13454de3b141399c96e2a","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":2159},"recordId":"SYN-PUB-02159"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Combining transcriptomic and proteomic measurements can, in principle, improve the accuracy of inferred gene regulatory networks by providing complementary information about regulatory state. We investigated whether joint modelling of RNA-seq and label-free quantitative proteomics data improves Bayesian network structure recovery in salt-stressed soybean, using a dataset of 20 paired samples. Three integration strategies were compared: early integration (concatenated feature matrices), late integration (separately learned networks merged by edge-weight thresholding), and a structured joint BGe scoring approach adapted for heterogeneous measurement types. The joint BGe approach yielded the highest F1 against the reference network (0.49 versus 0.41 and 0.38 for early and late integration, respectively), but required careful normalisation of cross-modality covariance terms. We detail the normalisation pipeline, identify failure modes when proteomics coverage is below 40% of transcriptomic gene set, and propose a practical quality filter for deciding when multi-omics integration is likely to help.","excerptSha256":"e8fdd5fcf215df56dbdefef9ab139debbe4fdc7aeae53cdea6adece6cfa20587","id":"ev-4550ffd9ce1aff7c81a1b419a73a59d566c5af54f083232e2aba84f9fe39ac4e","recordSha256":"75a4485ec5dffcfdb4ab98c36cad1788b17be68c97514b003418bcaa119c52cf","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":3328},"recordId":"SYN-PUB-03328"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0781","excerptSha256":"c8229d43fe347190149565eda439f3e6fe52c733342d687d5b1ea0ea059ed360","id":"ev-7cc21c30adf8dba3163ec7e3e680f32296a139d987a1e07840977d864dd8a8ca","recordSha256":"75a4485ec5dffcfdb4ab98c36cad1788b17be68c97514b003418bcaa119c52cf","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":3328},"recordId":"SYN-PUB-03328"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Ivona is actively looking for collaborators who generate high-throughput biological data (transcriptomics, proteomics, metabolomics) and need principled probabilistic approaches to network inference; she is less interested in purely applied software development roles. Her group has access to a 128-core HPC cluster at BioSense and can contribute compute time to collaborative projects.","excerptSha256":"af549041e673fee88e45b6b029866e875ebe5aab92ffd2e098f09c7a70c302f3","id":"ev-0ce56266b94d9895c9829e3cffb8365d81700469b1193e1e2d056538093c9ee9","recordSha256":"39166eb2bbbbc7ff16474c84a934f4870dba84e84304deb21b9446541ecf1d04","source":{"derivedRecordId":false,"fieldPointer":"/notes","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/780"},"recordId":"R0781"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"joint projects with companies processing plant or agricultural omics data\",\"contract research on probabilistic graphical model development\",\"consulting on Bayesian network methodology for sparse biological datasets\",\"student internships in computational bioinformatics\",\"co-supervision of industrial PhD candidates working on biological data integration\"]","excerptSha256":"fbbb0a78ffc02d8eb5afbc37205de6aa9b5a87e1390dca729bf47697bc7acd73","id":"ev-e138102ce5967908ddd43110ab0bfe3fae44bbddef6858c69f2e64bacca0e937","recordSha256":"39166eb2bbbbc7ff16474c84a934f4870dba84e84304deb21b9446541ecf1d04","source":{"derivedRecordId":false,"fieldPointer":"/open_to","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/780"},"recordId":"R0781"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Ivona has no formal industry employment history, though the STRESSNET project included an informal three-month knowledge exchange with a regional seed company to test the pipeline on proprietary stress phenotyping data under a non-disclosure agreement.","excerptSha256":"4d46f7a7bfb8c48a0a61c247124742761c9079bcf27b1a79f289e58c629bfad3","id":"ev-74c7bc66358a65d5ae059ece05fa683175c73762863cb8c78c8625f67314a3ea","recordSha256":"39166eb2bbbbc7ff16474c84a934f4870dba84e84304deb21b9446541ecf1d04","source":{"derivedRecordId":false,"fieldPointer":"/industry_experience","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/780"},"recordId":"R0781"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Standard cost-of-quality optimisation models derive an optimal inspection level by balancing appraisal and failure costs under assumptions of steady production volume. In high-mix low-volume assembly, batch sizes fluctuate widely and setup-to-run ratios are high, making these assumptions inappropriate. This paper formulates a cost model in which appraisal intensity (fraction of output subjected to each inspection stage) is chosen per batch given known batch size, product-specific baseline defect probability, and a per-unit inspection cost that includes setup amortisation. The model is solved analytically for single-stage inspection and numerically for two-stage inspection with in-circuit test followed by functional test. Using data from three product families (n = 847 batches over 18 months), the model yields appraisal intensity recommendations that differ from the company's fixed-sampling plan by −15% to +38% depending on batch size. Simulated application of the model reduces total quality cost per unit produced by an estimated 9.3% relative to the current fixed-intensity plan. Sensitivity analysis identifies defect probability estimation accuracy as the dominant uncertainty driver.","excerptSha256":"12144291c48ce33c93429a0cf5a1b165109789a2aa6d66253c6756759a603438","id":"ev-ff03ed59b0629ecae3bcc589649602e3baafbb0e8fe644246e6e92c5bc93f8ec","recordSha256":"b164b47353266e7824feed62576c9472af886e07b8bd9fbdabe715d37d82d696","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":9952},"recordId":"SYN-PUB-09952"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0945","excerptSha256":"a7619993b5ca313fd5a5c1420d62f5af29b47ad1668a8089ce2df6baa129c208","id":"ev-e0729c8443c593abed3077606767f7eb2dcee1b9308d38b1f806ea6c22649a1a","recordSha256":"b164b47353266e7824feed62576c9472af886e07b8bd9fbdabe715d37d82d696","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":9952},"recordId":"SYN-PUB-09952"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"When a contract electronics assembler takes on new product variants, inspection workload changes in ways that are difficult to predict from capacity calculations alone because defect rates, test durations, and re-test probabilities vary across variants. This paper describes a discrete-event simulation model of a three-stage inspection process (automated optical inspection, in-circuit test, functional test) calibrated with empirical process time distributions and per-variant defect rates collected over 12 months. The model is used to evaluate quality cost implications of four product-mix scenarios representing different ratios of high-complexity to standard-complexity boards. Results show that appraisal cost per good unit increases nonlinearly above a complexity ratio of approximately 0.45 due to bottleneck formation at functional test. Under the highest complexity scenario, total quality cost per unit is 23% above the baseline, with 71% of the increase attributable to internal failure costs generated by inspection queue-induced rework batching. The paper uses the simulation output to recommend a staffing and inspection sequencing policy that reduces this cost penalty to 11%.","excerptSha256":"793e0aa2564d3d48ec5bed6d3c29350c797d00e7d1f6fc809e3ffc8380fc191c","id":"ev-8a5f354a4ff2a009a2f0258edcf44a4e0965148d28fa183dfcf57881cbe79741","recordSha256":"f5b67b8f0e0a6be42e8f20251592f2f1262a0227d4596a91fa67c9d172bfa2eb","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":9954},"recordId":"SYN-PUB-09954"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0945","excerptSha256":"a7619993b5ca313fd5a5c1420d62f5af29b47ad1668a8089ce2df6baa129c208","id":"ev-34c5b32a28617ff144de45daff7e28063de761b08518033bd56469bf5ab1c8c6","recordSha256":"f5b67b8f0e0a6be42e8f20251592f2f1262a0227d4596a91fa67c9d172bfa2eb","source":{"derivedRecordId":false,"fieldPointer":"/authors/3/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":9954},"recordId":"SYN-PUB-09954"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"This paper presents a structured cost-of-quality (CoQ) data collection exercise conducted at a contract electronics manufacturer in Vojvodina producing between 40 and 200 units per batch across roughly 120 active product variants. Using a modified PAF (prevention–appraisal–failure) framework, the study captured 14 months of expenditure records and reclassified 23 cost categories previously buried in overhead accounts. Prevention costs accounted for 18% of total quality costs, appraisal for 41%, and internal failure for the remaining 41%, with external failure costs near zero due to strict outgoing inspection. Regression analysis on monthly data showed a statistically significant negative relationship between appraisal spending and internal failure costs (R² = 0.61), but only above a threshold appraisal intensity of roughly 2.3% of turnover. Below that threshold, increased inspection had no measurable effect on defect escape rates. The findings suggest that the company's inspection resources were concentrated at final test while upstream in-circuit test capacity was insufficient. Recommendations for rebalancing inspection effort are provided, along with a simple spreadsheet model validated against the 14-month dataset.","excerptSha256":"e10055220d9abe264a09094f97b5b6578e5dea0903fbbbcf98ade4cafb05e5b7","id":"ev-4e149cab38b92bb21d47c0009447a6ed29b46bf3a201d9709b3035c7136eaba3","recordSha256":"d22a34cf059fa94328280f43212199ea14b560a9355c5e300581472f56fcd355","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":9950},"recordId":"SYN-PUB-09950"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0945","excerptSha256":"a7619993b5ca313fd5a5c1420d62f5af29b47ad1668a8089ce2df6baa129c208","id":"ev-0f221457cc34d8e05709bb4cb4cd6b8fb3607a8534c98a98b2ca091dad02c7f0","recordSha256":"d22a34cf059fa94328280f43212199ea14b560a9355c5e300581472f56fcd355","source":{"derivedRecordId":false,"fieldPointer":"/authors/1/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":9950},"recordId":"SYN-PUB-09950"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"cost of quality\",\"appraisal–failure trade-off\",\"high-mix low-volume assembly\",\"hidden quality costs\",\"Pareto defect analysis\",\"electronics manufacturing\",\"prevention–appraisal–failure model\",\"defect reduction planning\",\"statistical process control\",\"quality cost optimisation\"]","excerptSha256":"a9e492f9f4268b069696c1b9d8f6413165ea59008ce1289cbb7e79f2fd287b98","id":"ev-902f3f0c12ff61b3af37efe7f6afe93ef1e9bc6d798c460c02632563359c77c7","recordSha256":"6a2c7dfd65cda30657c3bf3698f2bf7c42d8b67614f0678ef12a6c733c4ac2b8","source":{"derivedRecordId":false,"fieldPointer":"/keywords","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/944"},"recordId":"R0945"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0945","excerptSha256":"a7619993b5ca313fd5a5c1420d62f5af29b47ad1668a8089ce2df6baa129c208","id":"ev-ef5f4486c5e04633dea201128a573f25a8664456f41da2b0665cb83c9ab4b5c6","recordSha256":"6a2c7dfd65cda30657c3bf3698f2bf7c42d8b67614f0678ef12a6c733c4ac2b8","source":{"derivedRecordId":false,"fieldPointer":"/id","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/944"},"recordId":"R0945"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"The conventional PAF model accumulates failure costs by financial value without weighting for downstream consequence severity. A tombstoning defect detected at final test costs roughly the same in rework labour as a bridging defect, yet the bridging defect may carry higher field-failure risk if it escapes to the customer. This paper proposes an augmented PAF formulation in which each failure cost event is multiplied by a severity weight derived from an FMEA-style consequence scale anchored to field return data. The model is parameterised using 26 months of defect, rework, and field return records from a surface-mount technology line producing industrial sensor modules. Incorporating severity weights shifts the Pareto ranking of defect categories: solder voids on BGAs rise from fourth to first priority, while tombstoning falls from first to third. Re-prioritising corrective investment according to the severity-weighted ranking reduces estimated expected field failure cost by 17% relative to the unweighted ranking, at the same total corrective expenditure. The paper discusses the sensitivity of the results to severity score uncertainty and provides a worked numerical example.","excerptSha256":"01b6401ebf30ac708666297abbe1962f9414cb21024fbd8cccf71790a384efff","id":"ev-51fad9926b4f3e8e28374c0b2d1daa9dd9e04b45c8d1819253d7b8d955c6cdc2","recordSha256":"f42f6d71b588e061051d4d94d1ecefb7e19c865af05b43a54288a9dab618795a","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":7070},"recordId":"SYN-PUB-07070"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0945","excerptSha256":"a7619993b5ca313fd5a5c1420d62f5af29b47ad1668a8089ce2df6baa129c208","id":"ev-ae9a9569fdb42126b42b7322f04c57499bf624012f5211281c96b4e3954eea82","recordSha256":"f42f6d71b588e061051d4d94d1ecefb7e19c865af05b43a54288a9dab618795a","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":7070},"recordId":"SYN-PUB-07070"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Optimal Appraisal Intensity Under Batch Size Variability: A Cost Model for HMLV Electronics Assembly","excerptSha256":"2d9aaf9ef54221b603101071b7c7e7b7c5b6dabb0521f4bcb79c2da43a39b46e","id":"ev-0e9e1736d60fb96535cd4bf3c85f25d3ac728a70eeacda3a217a70c2842e355c","recordSha256":"b164b47353266e7824feed62576c9472af886e07b8bd9fbdabe715d37d82d696","source":{"derivedRecordId":false,"fieldPointer":"/title","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":9952},"recordId":"SYN-PUB-09952"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Mapping Prevention and Appraisal Expenditures in a Small-Batch PCB Assembly Line: A Case-Based Cost-of-Quality Analysis","excerptSha256":"2f9978defcdd65750b7c635b42bca31c194941fbe559f7cac26e5cb12819e5ce","id":"ev-be70f61a57c979747d712aa65b4194b9658be391e42f2cd48bb992660e6e5bd3","recordSha256":"d22a34cf059fa94328280f43212199ea14b560a9355c5e300581472f56fcd355","source":{"derivedRecordId":false,"fieldPointer":"/title","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":9950},"recordId":"SYN-PUB-09950"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Despite decades of guidance from ISO and industry bodies, formal quality cost reporting remains uncommon in small and medium-sized enterprises. This paper reports findings from a structured survey of 38 electronics manufacturers and contract assemblers in Serbia conducted in 2017–2018. Questionnaire sections covered CoQ measurement practices, cost categorisation schemes used, reporting frequency, and integration with management accounting. Thirty-one percent of respondents collected any formalised quality cost data; of those, 83% tracked only visible failure and scrap costs and none systematically estimated hidden costs. Ordinal logistic regression on firm-level predictors found that ISO 9001 certification was positively associated with CoQ data collection (OR = 3.2, p < 0.05), whereas firm size and export orientation were not significant after controlling for certification status. Interviews with quality managers at six firms identified lack of accounting system integration and perceived measurement effort as the primary barriers. The paper concludes with a compact CoQ implementation roadmap calibrated to the resource constraints typical of Serbian electronics SMEs.","excerptSha256":"6b20ac0cdb0d7deb8f36e98471ba60266fb30fb2f0542bb876dad098b88fe77c","id":"ev-354fc65185cedf815009bde81f6fec0434b2a4316717d7022efb1de9b5341eac","recordSha256":"c1c672fec4274cefde04f0e341bbd19db5fe7affca9b5b5c4e84bf938ec94b52","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":9953},"recordId":"SYN-PUB-09953"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0945","excerptSha256":"a7619993b5ca313fd5a5c1420d62f5af29b47ad1668a8089ce2df6baa129c208","id":"ev-70697d0b3ac96b1bd7049febab2a55d0ce738321b9ec4b436138e4749c9e9635","recordSha256":"c1c672fec4274cefde04f0e341bbd19db5fe7affca9b5b5c4e84bf938ec94b52","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":9953},"recordId":"SYN-PUB-09953"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"appraisal optimisation\",\"batch size variability\",\"inspection cost model\",\"HMLV\",\"two-stage inspection\"]","excerptSha256":"03c35fe44e91121ad8fe475f084f41e6140d8cbc50ca639def9514dc2b09210d","id":"ev-ddc8d6c7a20342c1d5047e65145bd6b41e5c29d3583f69ae12e663b2171b915c","recordSha256":"b164b47353266e7824feed62576c9472af886e07b8bd9fbdabe715d37d82d696","source":{"derivedRecordId":false,"fieldPointer":"/keywords","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":9952},"recordId":"SYN-PUB-09952"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"This chapter surveys quantitative cost-of-quality modelling approaches as applied to electronics assembly, tracing development from early PAF accounting frameworks through activity-based and process-cost models to more recent optimisation formulations. The authors examine the data structures required by each model class, practical obstacles to data collection in contract manufacturing, and the degree to which each approach has been validated against real production data rather than hypothetical examples. Particular attention is given to the treatment of batch-size variability, multi-stage inspection architectures, and the estimation of hidden failure costs—three characteristics that distinguish electronics contract manufacturing from the mass-production settings where most CoQ theory was originally developed. The chapter identifies several open problems: the absence of longitudinal studies tracking CoQ outcomes after model-guided interventions, limited integration with modern defect-traceability systems, and weak linkage between CoQ models and supply chain quality costs upstream of assembly. Directions for further empirical and modelling research are outlined.","excerptSha256":"779d91b4625b7ffe98982edc66cea52bdde3a9327eedb218fc20ed0c77983698","id":"ev-2708cc54cf1556a07f62182b2bc0f5fbe0b6f4a8f98aec5836c3988a35543b8f","recordSha256":"d8b9c5ed00498236667eda7c6c62456394297da5fc8ada3c504a8d6275817685","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":9955},"recordId":"SYN-PUB-09955"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0945","excerptSha256":"a7619993b5ca313fd5a5c1420d62f5af29b47ad1668a8089ce2df6baa129c208","id":"ev-9a8852692ce3a40196b10af299c79881862be26bd51554cf53f824ca6305f6a1","recordSha256":"d8b9c5ed00498236667eda7c6c62456394297da5fc8ada3c504a8d6275817685","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":9955},"recordId":"SYN-PUB-09955"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Quality Cost Reporting Practices in Serbian Electronics SMEs: A Survey of 38 Companies","excerptSha256":"9cd43328f824a055a1dcd4962c40530c9523b0a5828877627345e5abb889cc94","id":"ev-9061ad845aea200c2213c0a99a139c85591302f9fb23e876f8ac8fdba69e21fb","recordSha256":"c1c672fec4274cefde04f0e341bbd19db5fe7affca9b5b5c4e84bf938ec94b52","source":{"derivedRecordId":false,"fieldPointer":"/title","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":9953},"recordId":"SYN-PUB-09953"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Estimating Hidden Internal Failure Costs in Electronics Contract Manufacturing: A Structured Interview and Time-Study Approach","excerptSha256":"0027e27281780f52a31861bd3652ee4224e201e8eb6227f8dc9eb3a26413beee","id":"ev-944927d237fb5f3770144e66a4d3c2b540697ec88037809e6fa1891baf9ccee5","recordSha256":"b627130eda3865321b0313c8109b7fd10fbc31f21710534a551e93940c90cd04","source":{"derivedRecordId":false,"fieldPointer":"/title","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8789},"recordId":"SYN-PUB-08789"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0945","excerptSha256":"a7619993b5ca313fd5a5c1420d62f5af29b47ad1668a8089ce2df6baa129c208","id":"ev-c5a64f2bef8877aec9bf5eb42e7e93ae3d73d9b2598866869640a70102ba1b7a","recordSha256":"b627130eda3865321b0313c8109b7fd10fbc31f21710534a551e93940c90cd04","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8789},"recordId":"SYN-PUB-08789"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Bogićević prefers industry partners who can provide access to defect and rework records for at least 12 months; he is not in a position to take projects that require him to sign broad IP assignment agreements covering academic outputs. He is currently co-supervising two PhD students under HMLVCOST and has limited capacity for additional large project commitments before mid-2026.","excerptSha256":"6c58eea9f55d7ec6ae150c1ee8807990ec65c0005562306d8a8e2d5748a19ef5","id":"ev-988f95fecd58a451f14e03e99f8aea101e4ea082962c3c668d1d77846579b837","recordSha256":"6a2c7dfd65cda30657c3bf3698f2bf7c42d8b67614f0678ef12a6c733c4ac2b8","source":{"derivedRecordId":false,"fieldPointer":"/notes","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/944"},"recordId":"R0945"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"joint projects with electronics manufacturers on quality cost measurement and reduction\",\"contract research: CoQ baseline studies, defect Pareto analysis, inspection policy review\",\"industrial PhD supervision for candidates employed in electronics or contract manufacturing companies\",\"student internships in partner company quality departments\",\"consulting on PAF model implementation and quality cost reporting system design\"]","excerptSha256":"9c2e427f3d75cab45be99e169cfc49fac95cf9ed7f00a8f3b8623992f49d0185","id":"ev-3f40551e5be7984a0b278b1e6dd54e399b75ca66faca85a53e1fc9375c829225","recordSha256":"6a2c7dfd65cda30657c3bf3698f2bf7c42d8b67614f0678ef12a6c733c4ac2b8","source":{"derivedRecordId":false,"fieldPointer":"/open_to","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/944"},"recordId":"R0945"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Bogićević has conducted on-site data collection and consulting engagements at four electronics contract manufacturers in Serbia and one in Hungary, accumulating practical familiarity with SMT lines, AOI and ICT systems, and the quality management constraints typical of regional SMEs. He has delivered internal training on quality cost measurement for quality engineers at two partner companies.","excerptSha256":"19051d8349c806388ea130d8a5fd8ce3022d981ddfbbf7424f06fcc63246e8a1","id":"ev-d4b83dd3df5e34b3ea8538f19c72f91bd8ce1b93f27e3c7ef83022933c25b732","recordSha256":"6a2c7dfd65cda30657c3bf3698f2bf7c42d8b67614f0678ef12a6c733c4ac2b8","source":{"derivedRecordId":false,"fieldPointer":"/industry_experience","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/944"},"recordId":"R0945"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Three genotyping methods were benchmarked for their ability to discriminate between a production strain and spontaneous variants arising during fermentation of Bacillus subtilis for surfactin production. Thirty isolates from production seed banks at three facilities, along with 12 laboratory-generated spontaneous variants, were characterised. RAPD-PCR resolved clusters at the facility level but failed to separate closely related spontaneous variants in 4 of 12 cases. AFLP provided higher resolution but required two working days per sample set. Whole-genome SNP calling from Illumina short reads correctly identified all variants and provided information on the location and predicted functional impact of mutations. Sequencing cost per isolate was estimated at €38 in 2017 prices in a batch of 24 samples, already competitive with AFLP for batches above 16 isolates. Practical recommendations for selecting the appropriate method based on facility throughput and technical capacity are provided.","excerptSha256":"4d4fc2e3e9fe7f925d7da941e455a705c89e82afe842d29739ce58c75946e907","id":"ev-c0f48a60a3a50b7d5ac34e4d8b34af8fa4be4e41d332df540ec569f1e0313c2c","recordSha256":"23219ecb4ac15843ea7a073a158af469278fbef9573429c10e5f3a29caea487b","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":10447},"recordId":"SYN-PUB-10447"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R1005","excerptSha256":"e28858be3ed6f932dd0c4339b901cbf69d339e2341a83ee166f28a56f8f06bc4","id":"ev-26e0cbd65308b2d7ce9b2846c7cf57525c53d406834492f73a36e2b9cb2a68da","recordSha256":"23219ecb4ac15843ea7a073a158af469278fbef9573429c10e5f3a29caea487b","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":10447},"recordId":"SYN-PUB-10447"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Twelve consecutive 200 L pilot-scale batches of Bacillus licheniformis NRRL B-1365 were operated under nominally identical conditions to establish metabolic reproducibility benchmarks for α-amylase production. Broth samples at 24, 48, 72, and 96 h were analysed by LC-QTOF after methanol–water extraction. Of 214 assigned metabolite features, 38 showed inter-batch relative standard deviation below 15% and were designated candidate process fingerprint markers. Valine, leucine, and 2-oxoglutarate concentrations at 48 h were the strongest predictors of final α-amylase activity (R² = 0.81 in partial least-squares regression). Between-batch coefficient of variation in final titre was 9.3%, attributed primarily to dissolved oxygen trajectory differences resulting from impeller wear. The 38-metabolite fingerprint provides a practical in-process monitoring panel for detecting off-trajectory batches before harvest.","excerptSha256":"96dd4e0308fe4d13a268a1425ebc959282a4f4e0d36ac0072c2d8661f301d2d1","id":"ev-819b8cce86b4f7af3fb7735219ce0764092d1b49ec5eb2b116ca5b469ea42dba","recordSha256":"4571ebba36a53611fa393dfee4d172c5fe5eb78b3ec44b1e263ad713542a371c","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":10446},"recordId":"SYN-PUB-10446"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R1005","excerptSha256":"e28858be3ed6f932dd0c4339b901cbf69d339e2341a83ee166f28a56f8f06bc4","id":"ev-705dab63e3ee34c3f5663d17faa74f653388a96432fe71c120705012635955f4","recordSha256":"4571ebba36a53611fa393dfee4d172c5fe5eb78b3ec44b1e263ad713542a371c","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":10446},"recordId":"SYN-PUB-10446"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"A method is described for identifying deviating fermentation batches in real time by calculating a set of seven metabolite ratios from spent broth samples collected at defined process time points and comparing these ratios against control limits established from historical reference batches. The method applies to submerged bacterial fermentation processes producing extracellular enzymes or organic acids. It requires a single LC-MS analysis per time-point sample and produces a scalar deviation index that can trigger process intervention or batch rejection decisions without operator expertise in metabolomics data interpretation.","excerptSha256":"999d0a8aba9b61ce6c49a54a80fcdbd9b42d00ccb594e16ae3d7b226acfaca94","id":"ev-3908ca8b3be10fba720f9bf3c008eaea17994587bf1b0418d9a71cba3c976ea4","recordSha256":"4d061299839082eb1872e0f61237c3c2cea9c384a19762827c0f3e5461330911","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/patents.json","locator":{"format":"json","jsonPointer":"/patents/24"},"recordId":"SYN-PAT-025"},"sourceFileSha256":"3b67cedd20f63d29cdb2ee868bd11761f2b4ba7e562988cb4b2af466cefccfe0"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R1005","excerptSha256":"e28858be3ed6f932dd0c4339b901cbf69d339e2341a83ee166f28a56f8f06bc4","id":"ev-a46d5720ed8d67687c033c9697e48d62d71efcccad28595fa94f216c340559f6","recordSha256":"4d061299839082eb1872e0f61237c3c2cea9c384a19762827c0f3e5461330911","source":{"derivedRecordId":false,"fieldPointer":"/inventors/0/researcher_id","file":"data/patents.json","locator":{"format":"json","jsonPointer":"/patents/24"},"recordId":"SYN-PAT-025"},"sourceFileSha256":"3b67cedd20f63d29cdb2ee868bd11761f2b4ba7e562988cb4b2af466cefccfe0"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Six confirmed bacteriophage contamination events in a commercial Lactococcus lactis NZ9000 cheesemaking fermentation were retrospectively compared with 18 unaffected batches using archived broth samples from 4 h and 8 h post-inoculation. LC-QTOF metabolomics identified 22 metabolite features discriminating contaminated from unaffected batches at 4 h, before any detectable acidification lag was measurable by pH probe. Nucleotide pool metabolites (adenine, hypoxanthine, inosine) were consistently elevated in contaminated batches, consistent with phage-induced cell lysis releasing nucleic acid degradation products. A classification model built on seven metabolite ratios correctly assigned 23 of 24 batches in leave-one-out cross-validation. The study suggests that metabolomics-based early-warning monitoring could provide actionable contamination alerts one to two hours earlier than conventional pH-based detection.","excerptSha256":"94c3991ab1fcff550b8b5383fd2b48e961e33cbb079ee15165c0605fb50b6a8a","id":"ev-e1abb9ee3d839c99d8fe13e9a05b09db16301e287f307e46be7997acde85bcfc","recordSha256":"a004a7f99b05c3b88efc44c36c58722c89e76f0c02b68fcbef0b5c371f2afb61","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":10449},"recordId":"SYN-PUB-10449"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R1005","excerptSha256":"e28858be3ed6f932dd0c4339b901cbf69d339e2341a83ee166f28a56f8f06bc4","id":"ev-125f97a7cb5257e635a3db8f17d3194ae97955f068ee6089f663b62e4c81b272","recordSha256":"a004a7f99b05c3b88efc44c36c58722c89e76f0c02b68fcbef0b5c371f2afb61","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":10449},"recordId":"SYN-PUB-10449"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"A repeated-batch fermentation process for alkaline protease production using Aspergillus niger was operated over 12 consecutive batches in 5 L stirred-tank bioreactors. Protease activity in the harvest broth fell from 342 U/mL in batch 1 to 198 U/mL in batch 12, representing a 42% decline. Spent broth samples from each batch were profiled by HPLC-DAD for organic acids, amino acids, and sugar alcohols. Principal component analysis separated batches 1–4, 5–8, and 9–12 into distinct metabolic groups. Accumulation of gluconic acid (rising from 0.8 g/L to 4.1 g/L across batches) and concurrent depletion of free glutamate were the strongest discriminating variables. Spore morphology by light microscopy showed progressive conidiophore shortening correlating with gluconic acid concentration. The data suggest that acid accumulation combined with nitrogen recycling inefficiency drives the observed yield trajectory.","excerptSha256":"58ff8be7cb3f811abcd4231ec8b7af58eeb67959805045f6de20801df8c021e7","id":"ev-cb7174658531a4d220b09ab5f8e2db06ff923a0d9ade6b35b17600fcbff811c5","recordSha256":"2d85286f2aef8e83f4e04ba9653ac52519fd7ba2c7d73c114abfc827ff824aed","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6266},"recordId":"SYN-PUB-06266"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R1005","excerptSha256":"e28858be3ed6f932dd0c4339b901cbf69d339e2341a83ee166f28a56f8f06bc4","id":"ev-196277ba1bc5feac9bf213a7189b8366a7d30213cd04550c9cfa801014aea8cd","recordSha256":"2d85286f2aef8e83f4e04ba9653ac52519fd7ba2c7d73c114abfc827ff824aed","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6266},"recordId":"SYN-PUB-06266"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Plasmid retention in a recombinant Escherichia coli BL21(DE3) strain carrying a pET-based expression vector was monitored over 40-hour fed-batch cultivations at three different specific growth rates (0.10, 0.20, and 0.35 h⁻¹) without antibiotic selection pressure. Samples were taken every four hours and plated on selective and non-selective media to calculate plasmid retention fraction. At µ = 0.10 h⁻¹, retention remained above 92% throughout; at µ = 0.35 h⁻¹, retention fell below 70% after 28 hours. Flow cytometry using a chromosomally integrated GFP marker confirmed plasmid-free segregant fractions independently of plating. Volumetric productivity of the target protein (a recombinant chitinase) tracked plasmid retention closely. A simple population balance model was parameterised to the data and predicts retention behaviour within ±8% across the growth rate range tested.","excerptSha256":"71edc6cacd89c8c51e52ce312b09f24da06bce4887239ebacbe2ae1e895b5c9f","id":"ev-0043e9cec118d106599f06a5034bfb19598dfcf2487ede5217419030e2eababf","recordSha256":"43aac03719018454a9bfcd54426fdf3322970992a66efa1039cb691a6dbc0f20","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":10443},"recordId":"SYN-PUB-10443"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R1005","excerptSha256":"e28858be3ed6f932dd0c4339b901cbf69d339e2341a83ee166f28a56f8f06bc4","id":"ev-ba923cf9ff64d2b9b583cfcb123a4e8d90d698ef6587f8f12233f4c92f4c3404","recordSha256":"43aac03719018454a9bfcd54426fdf3322970992a66efa1039cb691a6dbc0f20","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":10443},"recordId":"SYN-PUB-10443"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"A 120-hour fed-batch fermentation of Corynebacterium glutamicum ATCC 13032 for glutamic acid production was operated with deliberate dissolved oxygen (DO) setpoint transitions (30%, 50%, and 70% air saturation, each held for 40 hours) to examine the effect of oxygen availability on both productivity and mutation accumulation. Glutamic acid titre was highest at 50% DO (48.3 g/L versus 41.1 g/L at 30% and 43.7 g/L at 70%). Whole-genome sequencing of 96 single colonies harvested from each DO phase identified mutation frequencies of 3.1 × 10⁻⁴, 1.8 × 10⁻⁴, and 5.6 × 10⁻⁴ mutations per genome per generation at 30%, 50%, and 70% DO respectively, with oxidative stress response genes overrepresented among mutations at 70% DO. The results caution against high-DO operation strategies adopted for productivity reasons without parallel strain stability monitoring.","excerptSha256":"4ec3a09dd4d4d903fb32fe4da0c103fe93855f7cae33e09588f0b30f232c93ee","id":"ev-f8987b35abbc2d29f60686110820d416847a486162525bc61e249059bc16f12e","recordSha256":"546666b18ca37f47bd47dc704ca6a6606c53d029da030cb62771bf6baa2d436f","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":10450},"recordId":"SYN-PUB-10450"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R1005","excerptSha256":"e28858be3ed6f932dd0c4339b901cbf69d339e2341a83ee166f28a56f8f06bc4","id":"ev-693b5354bb3891a77d6f3ee10b66277bd2a6cad4dcbd034aefca43d9797d3078","recordSha256":"546666b18ca37f47bd47dc704ca6a6606c53d029da030cb62771bf6baa2d436f","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":10450},"recordId":"SYN-PUB-10450"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Comparison of RAPD-PCR, AFLP, and whole-genome SNP calling for routine strain identity verification in industrial fermentation facilities","excerptSha256":"9b864c78bdd04d372081a6e7fafbec243f62e2ef84c52bebbdcd19f76b825335","id":"ev-737153d1f662f811ea05d5d344f93787b2edb21884d485032cc80092eca46924","recordSha256":"23219ecb4ac15843ea7a073a158af469278fbef9573429c10e5f3a29caea487b","source":{"derivedRecordId":false,"fieldPointer":"/title","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":10447},"recordId":"SYN-PUB-10447"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Stability of antibiotic-free plasmid retention in Escherichia coli BL21(DE3) under fed-batch conditions: influence of specific growth rate","excerptSha256":"06057f8f699fb28593f73f227252ef1c48257aa52dd79a98f1e3dac26442e12a","id":"ev-0454c97b1cb4e9e8f29ee3f1e50994ddf2529ec95995808469eed6c0d51bd6af","recordSha256":"43aac03719018454a9bfcd54426fdf3322970992a66efa1039cb691a6dbc0f20","source":{"derivedRecordId":false,"fieldPointer":"/title","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":10443},"recordId":"SYN-PUB-10443"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Repeated subculturing of Bacillus subtilis ATCC 6633 over 50 transfers was monitored by colony morphology scoring, sporulation frequency counting, and Illumina short-read whole-genome sequencing at transfers 10, 25, and 50. Sporulation frequency declined from 91% at transfer 1 to 67% at transfer 50. Sequencing identified seven single-nucleotide variants accumulating progressively, five mapping within the spo0A regulatory network. Plating-based morphology scoring correlated poorly with variant accumulation until transfer 35, suggesting that phenotypic methods underestimate early genomic instability. Glycerol cryostocks prepared at transfer 5 fully recovered wild-type sporulation and variant profiles after revival, supporting early banking as the primary mitigation strategy. The study provides quantitative thresholds for acceptable transfer number in inoculum propagation protocols used in industrial enzyme production.","excerptSha256":"a2d35489f9558991facf856b82caed7cd3fc733883c7d5158079eee5b5021584","id":"ev-01324a189ea3d070e1c1bad9a0821eda2def29cd57453079f89418a73d5d6154","recordSha256":"7c238c8b1520a3e5a3ceb87b23bc13f64012af7e08b72db3fedfc499d137ccae","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8218},"recordId":"SYN-PUB-08218"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R1005","excerptSha256":"e28858be3ed6f932dd0c4339b901cbf69d339e2341a83ee166f28a56f8f06bc4","id":"ev-b72dee887808ae20e052f3f9320e633709ebbbe14315391535c9d2634bdffc21","recordSha256":"7c238c8b1520a3e5a3ceb87b23bc13f64012af7e08b72db3fedfc499d137ccae","source":{"derivedRecordId":false,"fieldPointer":"/authors/1/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8218},"recordId":"SYN-PUB-08218"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"This book chapter presents a systematic framework for using multi-batch metabolomic profiling to distinguish strain-based from process-based causes of yield instability in submerged bacterial fermentation. The framework is structured in four phases: baseline fingerprinting across a minimum of eight representative batches, statistical control charting of key metabolite ratios, principal component monitoring for trajectory deviation, and targeted hypothesis testing for identified anomalies. A worked example using a Bacillus licheniformis β-glucanase fermentation across 22 consecutive batches at a pilot facility is used throughout. In the example, a statistically significant metabolite shift at batch 14 was traced to a medium preparation change rather than strain mutation, demonstrating the discriminatory power of the framework. Data handling procedures, minimum detectable effect sizes, and the number of metabolites required for reliable discrimination are discussed. The chapter is aimed at process microbiologists and bioprocess engineers with access to LC-MS instrumentation.","excerptSha256":"087913454e5f8d8cbf3d21943730c8288edf752c561c9060c31598ea0544978a","id":"ev-5527d83949f692d4386e124f450ab49e3b77b84f0dd8a1f57bcd1725210d9393","recordSha256":"9b3144e32bc86f6c99d14e027da834589e466533c1468541cc923ef818ff42aa","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":10453},"recordId":"SYN-PUB-10453"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R1005","excerptSha256":"e28858be3ed6f932dd0c4339b901cbf69d339e2341a83ee166f28a56f8f06bc4","id":"ev-cf0a40ca1c2728a84b2abc059159c5dc1b3506913d0da2f1498ab77394848e22","recordSha256":"9b3144e32bc86f6c99d14e027da834589e466533c1468541cc923ef818ff42aa","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":10453},"recordId":"SYN-PUB-10453"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"A commercial Corynebacterium glutamicum strain used in lysine production was revived from glycerol stocks stored at −80°C for periods of 1, 3, 5, and 8 years. After standardised revival and single-passage propagation, 200 colonies per stock age were characterised for growth rate on minimal medium, lysine secretion by HPLC, and cell morphology. Coefficient of variation in lysine titre per colony increased from 9% (1-year stocks) to 24% (8-year stocks). Mean titre per colony decreased by 11% over 8 years. Whole-genome sequencing of three low-titre and three high-titre colonies from each cohort identified no fixed mutations but showed varying proportions of a frameshift variant in the lysC gene encoding aspartate kinase. The finding indicates that long storage does not eliminate heterogeneity and that periodic single-colony isolation for stock renewal is warranted at intervals shorter than 5 years.","excerptSha256":"3c01bdc6258ea044a197c224ffd14b415aae87d44db37db8da1fa47446c928d2","id":"ev-c031d0872acb0479e5a8a7f770eab05ce89b94a99dd07b1a32e84240266c0e92","recordSha256":"6d6461ec803eb83faba7f21948fc1c28c184e27942883d9451cebae30f671c31","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":10444},"recordId":"SYN-PUB-10444"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R1005","excerptSha256":"e28858be3ed6f932dd0c4339b901cbf69d339e2341a83ee166f28a56f8f06bc4","id":"ev-9e90a553f008e0ba6eae1ecf734359e69b55461ee81bcc0da01d48d5926c22a9","recordSha256":"6d6461ec803eb83faba7f21948fc1c28c184e27942883d9451cebae30f671c31","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":10444},"recordId":"SYN-PUB-10444"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Prefers industry collaborations that involve access to genuine production data rather than model-system recreations; has limited availability for new projects in the first half of 2025 due to departmental accreditation obligations. Is particularly interested in connecting with NTP companies whose core product involves a fermentation step, even if microbiology is not their primary expertise.","excerptSha256":"744016c7f5f698531053052dae8c116540b613bfa9ded5b383df3e337aa873d2","id":"ev-df173a9151e9c950658e8856650f80e22124619a31fa25d7d386b1753c1ade3d","recordSha256":"42f37a3f0e464db744175ce4d4c0ee431d1e0adf9a26bf2daeb1fb36afb38510","source":{"derivedRecordId":false,"fieldPointer":"/notes","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/1004"},"recordId":"R1005"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"contract research on fermentation yield decline root-cause investigations\",\"joint projects with companies experiencing unexplained batch-to-batch variability\",\"industrial PhD supervision for candidates employed at fermentation companies\",\"consulting on seed-lot management protocol design and cell bank qualification\",\"student internships in fermentation metabolomics and strain characterisation\",\"equipment access to LC-QTOF and 5 L bioreactor suite for academic and industry partners\"]","excerptSha256":"6c13677e71fa5fc74ff2a8ec3f7a7515a29bf6cc6d6e55b1009913cda02b80de","id":"ev-5f19b5781feaee9953729a020cd98d365e4d928d764ba43f73ce7815c6afa976","recordSha256":"42f37a3f0e464db744175ce4d4c0ee431d1e0adf9a26bf2daeb1fb36afb38510","source":{"derivedRecordId":false,"fieldPointer":"/open_to","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/1004"},"recordId":"R1005"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Đukić has completed two industry-contract root-cause investigation engagements for regional fermentation manufacturers (one antibiotic producer, one enzyme producer) over the past eight years. He has also served as an external technical reviewer for a new product registration dossier at a Serbian pharmaceutical fermentation facility.","excerptSha256":"f1e7504ab2e0b6540c92d089f8fddfaf96da471afafbe4aafc191e53849f9c50","id":"ev-9c7dabc32d79c133f1e030e9253028dee3378739b395a3031b6f33b505876b50","recordSha256":"42f37a3f0e464db744175ce4d4c0ee431d1e0adf9a26bf2daeb1fb36afb38510","source":{"derivedRecordId":false,"fieldPointer":"/industry_experience","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/1004"},"recordId":"R1005"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"CNC cutting machines incur tool travel cost that depends on both the positions and the sequencing of piece outlines on the cutting plan. This paper addresses a bi-objective nesting problem minimising material waste and total cutting-path length simultaneously. A variable neighbourhood search (VNS) algorithm alternates between three neighbourhood structures: piece translation, piece swap, and sequence reversal of a cutting-path segment. The two objectives are combined via an ε-constraint method to generate approximate Pareto fronts. Experiments on six instances from an automotive gasket manufacturer show that layouts optimised solely for material utilisation have cutting paths 18 to 34% longer than layouts optimised for path length alone. VNS generates Pareto fronts of 15 to 22 non-dominated solutions per instance in under 20 minutes. A sensitivity analysis indicates that cutting-path savings of 12% can typically be achieved with less than 1 percentage-point sacrifice in material utilisation by selecting solutions in the knee region of the Pareto front.","excerptSha256":"92d19e7a870f1e1099b71be6093b3f2408ea0992d5e5da54cf512283e9a399c1","id":"ev-18e7be5b40c2bed37c0093c1bacd031381364a0c746fb3588deb4589b7d1a0d8","recordSha256":"3b8bfe508574d659463f4e9d8df9e442b30b41a0ba5e5f2fe8e41dbcc238b983","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8520},"recordId":"SYN-PUB-08520"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0775","excerptSha256":"ef80f02a35472638817f519a199264845b31bd82372d6beb7fe1b80f4d3fe300","id":"ev-4497959da20f4383975841fa6a1869b3bbdace65cd0e9b5c9b734d62c1b85d8a","recordSha256":"3b8bfe508574d659463f4e9d8df9e442b30b41a0ba5e5f2fe8e41dbcc238b983","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8520},"recordId":"SYN-PUB-08520"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"In garment production, adjacent cutting plans laid on the same spreading table ideally consume pieces from the same fabric roll to reduce changeover time and shade variation. This paper formulates roll-to-plan assignment and plan sequencing as a combined optimisation problem and solves it with a beam search procedure guided by a composite objective weighting material utilisation (70%) and roll changeover frequency (30%). The beam width is varied from 5 to 50 and evaluated on planning instances with 15 to 60 cutting plans drawn from production records of two Serbian garment manufacturers. At beam width 20, the procedure reduces average changeover frequency by 27% relative to first-fit sequencing while reducing material utilisation by only 0.4 percentage points. Solution time is under 4 seconds for all instances. The paper also reports a sensitivity analysis showing that the 70/30 objective weighting is robust within the range 60/40 to 80/20 for the tested instances.","excerptSha256":"5db3984eaaf4578e3cb32b3a63291874a54cad1ec4767a89da6a6c3056782515","id":"ev-51ebe8f2a4bf77fde52f91c5af11a88a1bc6d323e28af04ba1fd6aa0b77bff2f","recordSha256":"c4a96b8cf07b660249cd243c6ae6f7da091e484b5c8a981cd211c5e685deca66","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8516},"recordId":"SYN-PUB-08516"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0775","excerptSha256":"ef80f02a35472638817f519a199264845b31bd82372d6beb7fe1b80f4d3fe300","id":"ev-c29de80c0fca77f39df00635a48776086f978a7dcceca3a86213b066e1efdf29","recordSha256":"c4a96b8cf07b660249cd243c6ae6f7da091e484b5c8a981cd211c5e685deca66","source":{"derivedRecordId":false,"fieldPointer":"/authors/1/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8516},"recordId":"SYN-PUB-08516"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Furniture upholstery cutting presents a combined order-batching and nesting problem: grouping production orders into cutting batches determines which piece sets must be nested together, directly affecting fabric utilisation. This paper studies a medium-sized furniture manufacturer in Vojvodina with 80 to 120 active orders per week and 14 fabric types. A two-stage optimisation framework is proposed: an order-batching model based on set-partitioning assigns orders to batches sharing the same fabric type and colour, and a simulated annealing nesting solver then generates the physical layouts. The interaction between batch size and nesting efficiency is characterised over eight weeks of production data: increasing average batch size from 6 to 14 orders raises nesting utilisation from 76.3% to 81.7% at the cost of an average order lateness increase of 0.8 days. A Pareto front illustrating this trade-off is presented, and a simple weighting rule for practitioners is proposed. The framework was piloted over four weeks, reducing average fabric waste from 22.1% to 17.4%.","excerptSha256":"8531fe29364e05010859cc046d0a278a8fee80396d154bba27dd923b98fafcb4","id":"ev-a8e07a87cb1c1c8cf9032bb04625ccc11d427deaa128bcdade71da7bfd94c5aa","recordSha256":"05cf581741d8d328db562a4ea29e2f88aac61b90ea653596523958de6d2bc13e","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8518},"recordId":"SYN-PUB-08518"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0775","excerptSha256":"ef80f02a35472638817f519a199264845b31bd82372d6beb7fe1b80f4d3fe300","id":"ev-7581cf3a67a59e961e38b8d1a5f663cce9708d893f4f3eca436fdd0d1961a246","recordSha256":"05cf581741d8d328db562a4ea29e2f88aac61b90ea653596523958de6d2bc13e","source":{"derivedRecordId":false,"fieldPointer":"/authors/3/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8518},"recordId":"SYN-PUB-08518"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"In high-volume garment production, fabric is spread in multiple layers before cutting, and rolls of different remaining lengths must be combined to cover the required spread length. Mis-matched roll combinations create tail-end fabric scraps that contribute significantly to total waste. This paper models the multi-layer spread scheduling problem as a set-covering integer program in which columns represent feasible roll combinations for a given required spread length. A Lagrangian relaxation of the coverage constraints is solved iteratively to obtain lower bounds; the relaxed subproblem decomposes into independent knapsack problems solvable in polynomial time. On instances with 20 to 60 active fabric rolls and 8 to 25 cutting orders, the Lagrangian bound closes 89% of the integrality gap on average, and an LP-rounding post-processing step produces near-optimal integer solutions. Applied to three months of production data from a Serbian shirt manufacturer, the method reduces tail-end waste by 14.3% relative to current manual planning.","excerptSha256":"45bedfddf0cb5ef90c404232a67ddd578a458d3250638ebd6078a0b6a5777c23","id":"ev-94d1c58ff25fe200d569861facaf51f80ad26b6b5f767022cd84a602fede522d","recordSha256":"e4e891604c5149a75291c1f5f4bf26a93e683aba14ab5818a4f80ab764165bc5","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8522},"recordId":"SYN-PUB-08522"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0775","excerptSha256":"ef80f02a35472638817f519a199264845b31bd82372d6beb7fe1b80f4d3fe300","id":"ev-8f9af68053349c01c017e1d1394088d9badd99dcec395727b7910aea44fe4e65","recordSha256":"e4e891604c5149a75291c1f5f4bf26a93e683aba14ab5818a4f80ab764165bc5","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8522},"recordId":"SYN-PUB-08522"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"A beam search procedure for sequencing cutting plans to minimise fabric roll changeover","excerptSha256":"6a75c1db72fe81f5d42dbf3624fe549f13f04935c85b6946a9e0c106e91e2623","id":"ev-7f8a844c4d27bb71d3bfcc19046d07109853a4e027689d66ccb0a3171ad234a9","recordSha256":"c4a96b8cf07b660249cd243c6ae6f7da091e484b5c8a981cd211c5e685deca66","source":{"derivedRecordId":false,"fieldPointer":"/title","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8516},"recordId":"SYN-PUB-08516"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"beam search\",\"cutting-plan sequencing\",\"roll changeover\",\"apparel production\"]","excerptSha256":"46279ab4c3efa533b540b49bfa0699ca04277a1a0af68a44b3f5eecb8cc219d9","id":"ev-896c34716b67ccd6eb269534d79c66a8d92f07b20c6c2dd78534ae49d5a52ca8","recordSha256":"c4a96b8cf07b660249cd243c6ae6f7da091e484b5c8a981cd211c5e685deca66","source":{"derivedRecordId":false,"fieldPointer":"/keywords","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8516},"recordId":"SYN-PUB-08516"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Deterministic demand forecasts used in shift scheduling ignore forecast uncertainty and can lead to under- or over-staffing when actual orders deviate from plan. This paper develops a scenario-based stochastic programme for a contract electronics assembly plant in Subotica with 310 employees. Fifty demand scenarios per week were generated from three years of historical order data using a bootstrap procedure stratified by product family. The two-stage stochastic programme determines a committed weekly shift structure in the first stage and allows limited worker reallocation across lines in the second stage. Solving with sample average approximation required an average of 22 minutes per weekly planning problem on a 16-core workstation. Out-of-sample evaluation over 52 test weeks showed a 6.1 percent reduction in total labour cost variance compared to deterministic scheduling, with mean cost within 1.4 percent of the deterministic optimum. Sensitivity to scenario count is reported for 10, 25, 50, and 100 scenarios.","excerptSha256":"69a04a3c0228b8fc5b684f12d126f7cec33f498b0d36d814ce9eea5b2c07e73d","id":"ev-83e69215e30cda78e618a885f0450b7d69a5eb4a724fc9fea6b5c444765b4c08","recordSha256":"c5ef9b2cd7ee32dd54939f9754b341261f97477cddccd051705278d295438d1e","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8524},"recordId":"SYN-PUB-08524"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0775","excerptSha256":"ef80f02a35472638817f519a199264845b31bd82372d6beb7fe1b80f4d3fe300","id":"ev-5f21752db4607ad4ef31910fa897c9d0b55d9992ed24634da2f70404bf59d68d","recordSha256":"c5ef9b2cd7ee32dd54939f9754b341261f97477cddccd051705278d295438d1e","source":{"derivedRecordId":false,"fieldPointer":"/authors/1/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8524},"recordId":"SYN-PUB-08524"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Radovan Janković received his Dipl.Ing. in Industrial Engineering from the Faculty of Technical Sciences, University of Novi Sad in 1998, followed by an M.Sc. in 2002 and a Ph.D. in 2007, both from the same institution. His doctoral thesis addressed heuristic and exact methods for irregular polygon nesting in textile and leather cutting. He was appointed Full Professor in 2018. His research group focuses on computational geometry methods for 2D irregular nesting, marker-making optimisation, cutting-plan scheduling, and operations research techniques for minimising material waste in batch manufacturing. He teaches courses in operations research, production planning, and combinatorial optimisation. The group uses custom C++ and Python solvers, No-Fit Polygon libraries, and interfaces with commercial CAD/CAM cutting systems available in the faculty's Manufacturing Systems Laboratory.","excerptSha256":"1b3e84576df6a6383b4bf41ab109bf37e3077c7f7b49bee13ed17fb9804b5403","id":"ev-9cc99f790daa4bd3ced8ae4645e1e767511f87feb3f9fe35842e7295e96a07f9","recordSha256":"054de0612971d027a9a143fd79e3581c34a8633c7af9020e12fb82043baef47d","source":{"derivedRecordId":false,"fieldPointer":"/bio","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/774"},"recordId":"R0775"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0775","excerptSha256":"ef80f02a35472638817f519a199264845b31bd82372d6beb7fe1b80f4d3fe300","id":"ev-5ff1c7fc2d1504e231fa947f6580b9d7c3981e99ee426cfd2344e02266cd71a7","recordSha256":"054de0612971d027a9a143fd79e3581c34a8633c7af9020e12fb82043baef47d","source":{"derivedRecordId":false,"fieldPointer":"/id","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/774"},"recordId":"R0775"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Constructive nesting heuristics require hand-crafted placement rules whose performance varies substantially across polygon types. This paper investigates whether a graph neural network (GNN) can learn a placement policy from solved instances, replacing the manual rule. Each partial layout is represented as a graph in which nodes encode piece geometric features (area, perimeter, convexity ratio) and edges encode spatial relationships between already-placed pieces and the candidate piece. The GNN is trained on 4,800 solved instances generated from the CEE benchmark dataset using a supervised imitation-learning objective. On a held-out test set of 400 instances, the GNN-guided heuristic achieves average utilisation of 82.1% compared to 79.4% for the best constructive baseline, at comparable computation time. Generalisation experiments on ESICUP instances not seen during training show a 1.2 percentage-point improvement over baseline, suggesting transferability. The approach requires approximately 14 hours of training on a single GPU; inference time per placement decision is under 3 milliseconds.","excerptSha256":"714243aefb3dabcae9d59621bb484efc9f32f00aabb70b637f7de755e5037798","id":"ev-906ea3b3ee0247093780212e05ed04d0e027fced45f73d14b77b585a527711ee","recordSha256":"dda5e4f64a1a5a918948a4d94aa04f7144e84ae2500adba46e3469b02f2d3f95","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8521},"recordId":"SYN-PUB-08521"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0775","excerptSha256":"ef80f02a35472638817f519a199264845b31bd82372d6beb7fe1b80f4d3fe300","id":"ev-ca31c736ca45ab607538fabcd7e0dc665d609f2d6058daf8500b980b3019ef29","recordSha256":"dda5e4f64a1a5a918948a4d94aa04f7144e84ae2500adba46e3469b02f2d3f95","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8521},"recordId":"SYN-PUB-08521"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Cutting-plan scheduling in textile manufacturing requires allocating pattern-piece layouts to fabric rolls while respecting roll length, colour lot constraints, and due-date windows for garment assembly. This paper proposes a mixed-integer linear programming (MILP) model that integrates roll assignment, layout sequencing, and spread-height decisions. The model is tested on data from a mid-sized Serbian garment manufacturer with 12-week planning horizons and up to 85 active orders simultaneously. A column generation approach decomposes the problem into a master scheduling problem and a subproblem equivalent to a weighted bin-packing instance. On instances with up to 40 orders, the column generation procedure closes 94% of the LP relaxation gap within 600 seconds. Material utilisation improved from 78.2% to 81.9% compared to the company's manual planning procedure, and roll inventory turnover increased by 11% over a three-month pilot. Limitations regarding shade-banding constraints are discussed.","excerptSha256":"734aa2930c6fccfe9a233b98403d0805252bf90d4a58f0aa3058a7faa5308c8c","id":"ev-b42f7b987245e1b1a4ad7d6d56ae94d22b2d0c671a7be9b68d32468087f5e15b","recordSha256":"13ae1ad1c001d5a244883c143f683b39ac8cd9b4014b0500d7ca42c48fe2dd50","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8511},"recordId":"SYN-PUB-08511"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0775","excerptSha256":"ef80f02a35472638817f519a199264845b31bd82372d6beb7fe1b80f4d3fe300","id":"ev-a6015162c061de593f8ba5643581923ad270e90e0eccb1a29e799a600f65363c","recordSha256":"13ae1ad1c001d5a244883c143f683b39ac8cd9b4014b0500d7ca42c48fe2dd50","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8511},"recordId":"SYN-PUB-08511"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"This paper addresses the two-dimensional irregular strip packing problem, in which a set of non-convex polygonal pieces must be arranged within a fixed-width strip so as to minimise total strip length. A simulated annealing (SA) metaheuristic is implemented using no-fit polygon (NFP) calculations to evaluate feasible placements. The cooling schedule follows a geometric reduction with an initial temperature calibrated on 200 random placement trials per instance. Benchmark instances from the ESICUP repository with 25 to 197 pieces are used; the algorithm achieves an average material utilisation of 83.4%, improving the best-known solutions on four out of twelve instances. Computational time on a single-core 2.4 GHz processor ranges from 18 to 340 seconds depending on instance size and polygon complexity. Sensitivity analysis shows that NFP pre-computation accounts for 62% of total runtime, motivating future work on approximate NFP methods. The implementation is validated against a commercial marker-making system on a real denim fabric dataset of 48 pattern pieces.","excerptSha256":"b49c320c971572a1b97a987589b2ab8358f0e9b0b89b2c582930b8b57709bfc1","id":"ev-bc92c83c1fc20705243d8743766d488418a605b4b759b732c8a21fe4260f75ee","recordSha256":"fa627793a1989f21781e84a412fca7d07a418c4d31d63b06ccffd97b41b97dc5","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8509},"recordId":"SYN-PUB-08509"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0775","excerptSha256":"ef80f02a35472638817f519a199264845b31bd82372d6beb7fe1b80f4d3fe300","id":"ev-29775e75b1559b5d4eb4dcaf06a66656e37c6b4d2cfb1b34eea7c7943f7eb45a","recordSha256":"fa627793a1989f21781e84a412fca7d07a418c4d31d63b06ccffd97b41b97dc5","source":{"derivedRecordId":false,"fieldPointer":"/authors/3/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8509},"recordId":"SYN-PUB-08509"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"The invention relates to a computerised method for assigning fabric rolls of heterogeneous remaining lengths to multi-layer cutting spreads in garment and textile production. The method accepts as input a list of available roll lengths, required spread lengths, and layer counts per cutting plan, and outputs a roll-to-spread assignment minimising total tail-end waste. The core procedure applies a Lagrangian relaxation of a set-covering formulation solved iteratively with subgradient updates. The system interfaces with standard inventory management data exports and outputs assignments in a format compatible with common spreading machine control software. Granted by the Intellectual Property Office of Serbia.","excerptSha256":"26ffe03ec2102e8c5ba1b5143f8bf2f399db10b5c28862c926a15b7d808f8bd0","id":"ev-5818758a94b4c6e25d9eacb0266405fcdfa4ab4346d5a72fc4162c0575e5f392","recordSha256":"89b3b2c1c66357e8fd2d4511051a1d79c487861d057720955c65e99e64a3b78f","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/patents.json","locator":{"format":"json","jsonPointer":"/patents/55"},"recordId":"SYN-PAT-056"},"sourceFileSha256":"3b67cedd20f63d29cdb2ee868bd11761f2b4ba7e562988cb4b2af466cefccfe0"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0775","excerptSha256":"ef80f02a35472638817f519a199264845b31bd82372d6beb7fe1b80f4d3fe300","id":"ev-0221b9290d846d30e49a3299527df4b9476badb25fa74a0412a094e15b86870e","recordSha256":"89b3b2c1c66357e8fd2d4511051a1d79c487861d057720955c65e99e64a3b78f","source":{"derivedRecordId":false,"fieldPointer":"/inventors/0/researcher_id","file":"data/patents.json","locator":{"format":"json","jsonPointer":"/patents/55"},"recordId":"SYN-PAT-056"},"sourceFileSha256":"3b67cedd20f63d29cdb2ee868bd11761f2b4ba7e562988cb4b2af466cefccfe0"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Janković is particularly interested in working with companies that have digitised their pattern libraries and can share polygon data; he is less suited to projects requiring physical prototyping or hardware development. He has supervised five completed Ph.D. theses and currently has capacity for one additional doctoral student.","excerptSha256":"b2b0243c61d7a567bf90d2cc09018b46044826ea5702a9b7e7199b1376b21082","id":"ev-380b88b059ecd07854bb4b609f83a632e9bbe40ac6a31c929e34d27ef2d110fe","recordSha256":"054de0612971d027a9a143fd79e3581c34a8633c7af9020e12fb82043baef47d","source":{"derivedRecordId":false,"fieldPointer":"/notes","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/774"},"recordId":"R0775"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"joint projects with companies on material waste reduction\",\"contract research on nesting and cutting-plan optimisation\",\"industrial PhD supervision\",\"student internships in computational geometry and operations research\",\"consulting on CAD/CAM nesting software selection and configuration\",\"access to the Manufacturing Systems Laboratory for cutting layout testing\"]","excerptSha256":"a284e339405d10a0123f0177b09a2a5ca0f4b4975bb9b292fa2eccbe92076872","id":"ev-796d82c5c81f91ccf46e0dcf9993864b61818cf0ca70271243f242847a305c22","recordSha256":"054de0612971d027a9a143fd79e3581c34a8633c7af9020e12fb82043baef47d","source":{"derivedRecordId":false,"fieldPointer":"/open_to","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/774"},"recordId":"R0775"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Janković has consulted for three Serbian garment manufacturers on nesting software procurement and production planning process redesign, most recently during the MATOPT project pilot phase in 2022–2023. He has served as a technical reviewer for a Serbian government procurement of CAD/CAM cutting systems.","excerptSha256":"39c06ddb3968447f87fe25583a78efa4aa97722cc6a274d5fb0247746a36f961","id":"ev-c7767534d82007237a37736a27ffda61f95b9fc8f8df0451eb4d3b8441666476","recordSha256":"054de0612971d027a9a143fd79e3581c34a8633c7af9020e12fb82043baef47d","source":{"derivedRecordId":false,"fieldPointer":"/industry_experience","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/774"},"recordId":"R0775"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Cloud providers offer preemptible compute instances at discounted prices, but tasks running on such instances may be interrupted when market demand rises above supply. This paper models a batch-processing workload that uses a mix of on-demand and preemptible instances as an M[X]/M/c queue with unreliable servers, where server failures correspond to preemption events governed by an exogenous price process modelled as a two-state continuous-time Markov chain. The stationary distribution is derived numerically using truncated state-space methods. A bidding strategy that sets the maximum acceptable spot price as a function of the remaining job deadline and estimated preemption rate is derived from the first-passage time distribution. Experiments use preemption rate estimates from a 30-day synthetic price series generated with parameters drawn from published empirical studies of spot markets. The bidding strategy achieves a 22 percent cost reduction over a fixed-bid baseline while maintaining deadline miss probability below 5 percent.","excerptSha256":"c157cd0916037af68873698de5275809f39f1310ece46aff6c53bcfb46c82439","id":"ev-25f78470b0924f7dad5b8dd81d211039600120b73beecb5119d3436834c523b0","recordSha256":"7a0069f9e4f352e108ce2f0bdb63cc4a3ac865114f944a95d48915062d3802c7","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6871},"recordId":"SYN-PUB-06871"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0595","excerptSha256":"b6754bf159957e65db19c202c3f0012fb4688d23c9ab86b325877ff9fada1587","id":"ev-8deb8262457648cac62260d37c3e86988f663f544dbeebf950e25ad1b3b4c72a","recordSha256":"7a0069f9e4f352e108ce2f0bdb63cc4a3ac865114f944a95d48915062d3802c7","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6871},"recordId":"SYN-PUB-06871"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Clinical risk scores are expected to behave monotonically with respect to certain predictors: higher creatinine should not decrease predicted renal failure risk, for example. Unconstrained gradient boosted trees frequently violate such expectations, reducing clinician trust. This study applied monotonicity constraints available in the XGBoost and LightGBM libraries to three clinical risk modelling tasks: 30-day hospital readmission (n=7,940), acute kidney injury within 48 hours of ICU admission (n=3,210), and postoperative pulmonary complication (n=2,880). Constraint specifications were derived from a literature review and validated with three clinicians. Constrained models showed AUPRC reductions of 0.01 to 0.03 relative to unconstrained models, while achieving full monotonicity compliance on all constrained features. A clinician survey conducted after model demonstration rated constrained models as significantly more trustworthy on a five-point Likert scale (mean 4.1 vs 3.0, p < 0.01, Wilcoxon signed-rank test). The trade-off between performance and interpretability was assessed using a proposed composite score.","excerptSha256":"ee23b27835b2691607d3bdcb03c3529a6805c895275d81cd79637d51070c3a0e","id":"ev-92e7aa64b2b58045274b6c4cfd7c8a5239693eea7d2c3e08c3b8c930dc8ead64","recordSha256":"d0715bf26dd3d7b6fb9df29a04626bac4e1d1a51f97826f2b8818dbf1003390b","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":5144},"recordId":"SYN-PUB-05144"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0595","excerptSha256":"b6754bf159957e65db19c202c3f0012fb4688d23c9ab86b325877ff9fada1587","id":"ev-e64f21347bf14c0fa3450599cddc831931351f903dcc7c601055a69fe0e8bb33","recordSha256":"d0715bf26dd3d7b6fb9df29a04626bac4e1d1a51f97826f2b8818dbf1003390b","source":{"derivedRecordId":false,"fieldPointer":"/authors/4/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":5144},"recordId":"SYN-PUB-05144"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Classification thresholds are routinely left at the default 0.5 in many applied clinical studies despite well-established evidence that asymmetric misclassification costs make this choice suboptimal. This paper reviews threshold selection methods including cost-minimisation, F-beta maximisation, utility-maximisation, and the Youden index, and provides an empirical comparison across twelve publicly deposited and in-house clinical tabular datasets ranging from 500 to 28,000 samples. Datasets cover six disease areas including cardiovascular risk, diabetes complications, and haematological malignancy. The comparison uses nested cross-validation to avoid threshold selection bias. Results show that cost-minimisation and F-beta methods are broadly equivalent when cost ratios are specified accurately, but that the Youden index consistently underperforms in high-imbalance settings with imbalance ratios above 1:15. The paper also provides guidance on eliciting cost ratios from clinical stakeholders and documents a structured interview protocol used with six clinicians at a teaching hospital in Novi Sad.","excerptSha256":"9963a79cb777eb1af6f5c32a06e28e3f8baffccf7e53e427b3c29d44d1f15a0c","id":"ev-843d5911e250a1c0401e4315cac79eee785d7ead466325dbc92e33ddfdc0497c","recordSha256":"cfa674862410a06ba5addeb94c09155a875ed4be0f6ab571a8ccdb613b487e8c","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6862},"recordId":"SYN-PUB-06862"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0595","excerptSha256":"b6754bf159957e65db19c202c3f0012fb4688d23c9ab86b325877ff9fada1587","id":"ev-c0f063a7aa3c99ed94a3a580b7754b8a375e92b2e83c4ac2e3574bc7467c403d","recordSha256":"cfa674862410a06ba5addeb94c09155a875ed4be0f6ab571a8ccdb613b487e8c","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6862},"recordId":"SYN-PUB-06862"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Many clinical laboratory studies involve datasets in which the number of measured analytes or derived features approaches or exceeds the number of patient records. In such settings, feature selection outcomes vary substantially across bootstrap samples of the training data, threatening reproducibility. We examined selection stability for five filter and three wrapper methods applied to six datasets with 80 to 340 samples and 30 to 210 features, all drawn from clinical biochemistry and immunology laboratory records. Stability was quantified with the Nogueira–Brown stability index. Mutual information filtering showed the highest stability index across five of six datasets. Wrapper methods based on recursive feature elimination with support vector machines showed strong predictive performance but low stability indices, with mean index values below 0.45. A two-stage procedure—filter pre-selection followed by wrapper refinement—improved both stability and cross-validated AUC in four datasets. Practical guidelines are provided for laboratories operating under sample-size constraints.","excerptSha256":"8bd7d867c927aa09673c3958b313da27b5c9380015495dba9b1524ec5dac8cb3","id":"ev-6e675ab0187353ebd4149e47d8c07ccf7f5ae92e85fbff9daf2e455d2fabcd0b","recordSha256":"cf7c9d3c729f3c96335521078e7cee4648eb7eb823ecd8471219aaef2ac63090","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6645},"recordId":"SYN-PUB-06645"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0595","excerptSha256":"b6754bf159957e65db19c202c3f0012fb4688d23c9ab86b325877ff9fada1587","id":"ev-89c2f0cc0970952db3c0d5f5a8bdabc0097bbcafc262ed040805b20b691ec0ce","recordSha256":"cf7c9d3c729f3c96335521078e7cee4648eb7eb823ecd8471219aaef2ac63090","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6645},"recordId":"SYN-PUB-06645"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"spot instances\",\"preemption\",\"unreliable servers\",\"bidding strategy\",\"first-passage time\"]","excerptSha256":"138cb5241c9d3aba6209bb09de603ee4bda38f231e2ee0759f53b2ba48d973aa","id":"ev-bbaaa29a15a99487f743de945f018c197276d1bcfd96b5313298e41393b39425","recordSha256":"7a0069f9e4f352e108ce2f0bdb63cc4a3ac865114f944a95d48915062d3802c7","source":{"derivedRecordId":false,"fieldPointer":"/keywords","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6871},"recordId":"SYN-PUB-06871"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"This book chapter provides a self-contained introduction to cost-sensitive classification intended for laboratory informaticians and biostatisticians with limited background in machine learning. It covers the formulation of asymmetric loss functions, the construction of cost matrices from clinical outcome data, threshold-moving as a post-hoc strategy, and direct cost-minimising training through modified learning objectives. Each concept is illustrated with worked examples drawn from haematology, microbiology, and clinical chemistry laboratory settings. The chapter also discusses common pitfalls, including the use of accuracy as a performance metric when classes are imbalanced, the conflation of probabilistic calibration with discrimination, and the failure to propagate cost assumptions into cross-validation. Pseudocode is provided for the main algorithms. The target audience is practitioners who need to communicate method choices to clinical colleagues and regulatory reviewers.","excerptSha256":"357856f3bb0363196983f7cc0084868a71773c4a0379f10eb262fa0e1731ae8f","id":"ev-2fbabb4d15d3f387f45d08274eb3263bc3e53b2d5a794a2b44ed02adbf08d380","recordSha256":"468df7f9ee16c0f0000c4ba8a7ebdc89787ff98058cb12f6a918f061cc2567de","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6864},"recordId":"SYN-PUB-06864"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0595","excerptSha256":"b6754bf159957e65db19c202c3f0012fb4688d23c9ab86b325877ff9fada1587","id":"ev-c4c2f5a995ae43800f03ed17ce86d80ef2ef290285bec6413054b2eea8c22520","recordSha256":"468df7f9ee16c0f0000c4ba8a7ebdc89787ff98058cb12f6a918f061cc2567de","source":{"derivedRecordId":false,"fieldPointer":"/authors/1/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6864},"recordId":"SYN-PUB-06864"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"CLASSLAB addressed the gap between classification algorithm research and practical laboratory deployment by developing a modular software toolkit for cost-sensitive threshold optimisation and model calibration. The project collected retrospective laboratory datasets from three hospitals in Vojvodina under data-sharing agreements and used them to benchmark methods across a range of imbalance ratios and cost structures. A secondary objective was to develop standardised protocols for eliciting misclassification cost estimates from clinical staff. The toolkit was implemented in Python and released as open-source software. The project supported two doctoral students and produced seven peer-reviewed publications. User testing with laboratory informaticians at the partner hospitals informed two revision cycles of the software interface.","excerptSha256":"50284655dda951c4e71520c4125c68824d24ff744f233153a7db5fe2924d1bbf","id":"ev-bc75addd35dd2be5da2f9ebd57e73442624d8f518c27a82371e37e7dcf2ebab3","recordSha256":"51fc793a9b12c16a40081dffefb9fa70fd3ab403cbab4731f56a21cd293cbc7e","source":{"derivedRecordId":false,"fieldPointer":"/summary","file":"data/projects.json","locator":{"format":"json","jsonPointer":"/projects/247"},"recordId":"SYN-PRJ-0248"},"sourceFileSha256":"847cb13091f91ed9fe57f6aea0088cd1e66ed0263c86a5090f169a2b30ab3a1a"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0595","excerptSha256":"b6754bf159957e65db19c202c3f0012fb4688d23c9ab86b325877ff9fada1587","id":"ev-79c4b52495679d4d11a8f821cc1db2aaa96d084c7e89c249b52f64dc529e9bd8","recordSha256":"51fc793a9b12c16a40081dffefb9fa70fd3ab403cbab4731f56a21cd293cbc7e","source":{"derivedRecordId":false,"fieldPointer":"/participants/0/researcher_id","file":"data/projects.json","locator":{"format":"json","jsonPointer":"/projects/247"},"recordId":"SYN-PRJ-0248"},"sourceFileSha256":"847cb13091f91ed9fe57f6aea0088cd1e66ed0263c86a5090f169a2b30ab3a1a"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Thyroid function is routinely assessed through a panel of three to five serum markers. Clinical guidelines define five functional categories—overt hypothyroidism, subclinical hypothyroidism, euthyroid, subclinical hyperthyroidism, and overt hyperthyroidism—but borderline cases generate significant inter-clinician disagreement. We trained one-vs-one kernel SVMs with radial basis function kernels on 6,730 labelled thyroid panels collected between 2012 and 2017 at two endocrinology outpatient clinics. Multiclass probability calibration used Platt scaling extended to the one-vs-one scheme. Overall accuracy on a stratified test split of 1,346 samples reached 88.3%. The two subclinical categories were most often confused, with confusion concentrated in samples where TSH differed from the midpoint of the reference range by less than 0.8 mIU/L. A confidence threshold was set such that panels with posterior probability below 0.65 for any predicted class were flagged for clinical review; this threshold captured 94% of misclassified cases at a flag rate of 21%.","excerptSha256":"34a8c837237df3a8d08a28c82f6ede6e0fa6454c97845cf282b85144869c0283","id":"ev-e78959fe5fcc33b9973c3b63647f83fd36b35eef68cb93f3046bfd252cd96e11","recordSha256":"e52dc628769e36ce144f2cafc90d76ead2eaec96b99d9deec90e19ff832d374f","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6863},"recordId":"SYN-PUB-06863"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0595","excerptSha256":"b6754bf159957e65db19c202c3f0012fb4688d23c9ab86b325877ff9fada1587","id":"ev-91a413d415079546c9b321aabb8627f7fed8e436d2d1ee4906480eaec7b76701","recordSha256":"e52dc628769e36ce144f2cafc90d76ead2eaec96b99d9deec90e19ff832d374f","source":{"derivedRecordId":false,"fieldPointer":"/authors/1/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6863},"recordId":"SYN-PUB-06863"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Threshold optimisation under asymmetric misclassification costs: a review and empirical comparison on clinical tabular datasets","excerptSha256":"2a6ab355ddb7ca909764c4ca8bc47c614890ae00836d081154432e6fd0dbdfff","id":"ev-15f6b731aff3f59bd8d045195b2d8569846663e7501e859b2e9d682d5ea2d627","recordSha256":"cfa674862410a06ba5addeb94c09155a875ed4be0f6ab571a8ccdb613b487e8c","source":{"derivedRecordId":false,"fieldPointer":"/title","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6862},"recordId":"SYN-PUB-06862"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"supervised classification\",\"imbalanced classes\",\"threshold optimisation\",\"clinical decision support\",\"multivariate clinical data\",\"cost-sensitive learning\",\"random forests\",\"laboratory informatics\",\"ROC analysis\",\"feature selection\"]","excerptSha256":"00c7d3c763169034e23175c22cb036de08b28f1e47313191cb84c9c62a4ad91e","id":"ev-7d0035b08f7216b9816df0e20e05062dd1f8557f320b7311e651ac49693eb13d","recordSha256":"14af32808e022b7220cadde4aa7cd9efbd6835825fd8772687d1f68ee252efd1","source":{"derivedRecordId":false,"fieldPointer":"/keywords","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/594"},"recordId":"R0595"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0595","excerptSha256":"b6754bf159957e65db19c202c3f0012fb4688d23c9ab86b325877ff9fada1587","id":"ev-c271f449562440721a00dbee808723123760cb64ce82cb34bb45ae9755cc4f65","recordSha256":"14af32808e022b7220cadde4aa7cd9efbd6835825fd8772687d1f68ee252efd1","source":{"derivedRecordId":false,"fieldPointer":"/id","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/594"},"recordId":"R0595"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Cost-Sensitive Classification Algorithms for Laboratory Decision Support in Resource-Constrained Settings","excerptSha256":"1e312b75ab7d6120093f08beb3108cca6e40298e1ddab048cf33cce9441871b1","id":"ev-394b72e4869179090f6501a992a3dfb2ac2784da4ba0220ee772f9223a01c5a0","recordSha256":"51fc793a9b12c16a40081dffefb9fa70fd3ab403cbab4731f56a21cd293cbc7e","source":{"derivedRecordId":false,"fieldPointer":"/title","file":"data/projects.json","locator":{"format":"json","jsonPointer":"/projects/247"},"recordId":"SYN-PRJ-0248"},"sourceFileSha256":"847cb13091f91ed9fe57f6aea0088cd1e66ed0263c86a5090f169a2b30ab3a1a"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Hypoglycaemic episodes in insulin-treated type 2 diabetes patients are associated with significant morbidity but are difficult to predict from laboratory data alone. We constructed a two-source feature set combining 14 routine biochemical markers with 9 patient-history variables including insulin dose, meal regularity score, and prior episode count, recorded for 2,640 outpatients over 36 months. Prediction targets were hypoglycaemic episodes confirmed by glucose monitor readings below 3.9 mmol/L within a 72-hour horizon. Random forests, logistic regression, and naive Bayes were trained on 80% of the data and evaluated on the remaining 20%. The combined feature set improved F1 from 0.48 (laboratory-only) to 0.61 (combined) for random forests. Shapley additive explanation (SHAP) values identified prior episode count, HbA1c, and insulin dose as the three highest-magnitude contributors. Threshold was set at 0.38 rather than 0.5 to increase sensitivity, following the cost analysis agreed with the clinical partner.","excerptSha256":"ac0a4ab73e3e7a40cfc399b6a1df9d70862f55811448c093bc428c1a8412eba9","id":"ev-3527cd117a4692971cb90b44d5c2cd592205b63018f6376f8c35a5d5c70338e3","recordSha256":"ad6430f7615741008cac21c8132382a37d54600a8c41665b3f921239c6d3fda3","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6867},"recordId":"SYN-PUB-06867"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0595","excerptSha256":"b6754bf159957e65db19c202c3f0012fb4688d23c9ab86b325877ff9fada1587","id":"ev-1f6cf7a1529d31f3230f62656ee97ab34f85b97b1aaf392f4f0d7891d272e061","recordSha256":"ad6430f7615741008cac21c8132382a37d54600a8c41665b3f921239c6d3fda3","source":{"derivedRecordId":false,"fieldPointer":"/authors/3/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":6867},"recordId":"SYN-PUB-06867"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Kmeťová is on sabbatical for the spring semester of 2026 and is not taking on new teaching commitments during that period, but remains available for research collaboration and short consulting engagements. She has expressed a preference for partnerships that provide access to real laboratory datasets rather than purely algorithmic or theoretical collaborations.","excerptSha256":"515ebd54ead63b1affa92f38034eda489a77569a0616309b9ca6d728f047211e","id":"ev-05ab8c33063fb3c6a33ebcdf4b92357cb7eb84f05d37fabb057b63a079f4bce9","recordSha256":"14af32808e022b7220cadde4aa7cd9efbd6835825fd8772687d1f68ee252efd1","source":{"derivedRecordId":false,"fieldPointer":"/notes","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/594"},"recordId":"R0595"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"joint projects with companies developing laboratory information systems or diagnostic software\",\"contract research on classification algorithm development and validation for clinical data\",\"consulting on threshold selection, imbalanced-class handling, and model evaluation methodology\",\"industrial PhD supervision for candidates working on applied machine learning in laboratory or medical device contexts\",\"student internships (MSc level) involving real dataset analysis with clinical partners\"]","excerptSha256":"cb9bcaf41ab2853e212a80b93c5da430d0bc1be6d15b4cf6fb3c9502ef60d7de","id":"ev-c87ad6444006aa055f174a86ed95e7ec3c2c4b5ea4a3ace7351f584e00196680","recordSha256":"14af32808e022b7220cadde4aa7cd9efbd6835825fd8772687d1f68ee252efd1","source":{"derivedRecordId":false,"fieldPointer":"/open_to","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/594"},"recordId":"R0595"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Kmeťová provided technical consulting to a regional laboratory software vendor over an 18-month period (2020–2021), advising on the integration of automated flagging algorithms into their laboratory information management system. The engagement resulted in a co-authored technical report and an ongoing data-access agreement used in the CLASSLAB project.","excerptSha256":"993dad7c9e4f52a910bb6bcae2784e99ef26e5bd95a66cc040127b81ca06a533","id":"ev-86d3d56a080ffdfce0f0ed1b4cd1242050cda9b96f7d389b73ac3d8b1e4cd4f7","recordSha256":"14af32808e022b7220cadde4aa7cd9efbd6835825fd8772687d1f68ee252efd1","source":{"derivedRecordId":false,"fieldPointer":"/industry_experience","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/594"},"recordId":"R0595"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Machine breakdowns introduce capacity uncertainty that static lead-time quotation and pricing models ignore. This paper reports a discrete-event simulation study of a six-machine job shop with three job classes, modelled in Arena. Four pricing policies are compared: fixed price with first-come first-served sequencing; fixed price with shortest-processing-time priority; workload-triggered price steps; and a continuous price function of current queue length. Machine unavailability is modelled with Weibull-distributed time-to-failure drawn from maintenance records of a Vojvodina engineering company. Over 200 replications of a 90-day horizon, the continuous pricing policy achieves a 14.3% higher net revenue than fixed pricing while holding average customer waiting time approximately constant. The workload-step policy performs similarly but requires only three price tiers, suggesting practical implementability. Results are sensitive to the assumed demand elasticity; the paper conducts a parametric sweep over the range estimated in earlier stated-preference work.","excerptSha256":"6de54412d02142affe447fed5f2e0d6162d9c858e946c8cd7ef60a745fe5b5f6","id":"ev-e08dcde535da83936873bae36d529bf9ed0d3391458a9cb9a2bf6cf4b0b5d92e","recordSha256":"95e57d829160471fa9771e06f6f77b5eb82fcdacfeda63b5a0ea82b20c639d19","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":7222},"recordId":"SYN-PUB-07222"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0633","excerptSha256":"21ff234b9a3afd93e725b87c798514eeed3ef3801a425051d9ff69744522f498","id":"ev-07083e638a8b62605368bb23b359b912bb0b66c4158cb41d986606744dbe6425","recordSha256":"95e57d829160471fa9771e06f6f77b5eb82fcdacfeda63b5a0ea82b20c639d19","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":7222},"recordId":"SYN-PUB-07222"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"This paper develops a steady-state pricing model for a single-machine make-to-order shop operating under Poisson arrivals and generally distributed service times (M/G/1 queue). The model links the posted price of a job to the current workload level measured in machine-hours, allowing the firm to dampen demand when the queue is long and stimulate it when capacity is idle. A profit function combining contribution margin with holding and tardiness costs is maximised over a finite price grid. Numerical experiments calibrated on data from a small metal-fabrication firm in Vojvodina show that workload-contingent pricing reduces average tardiness by 18% and increases weekly contribution margin by 11% compared with a static price policy. Sensitivity analysis reveals that the benefit of dynamic pricing grows with demand variability and diminishes when the coefficient of variation of service times exceeds 2.0. The paper discusses implementation requirements, including real-time workload visibility and customer acceptance of variable quotes.","excerptSha256":"1e3c5cfe65f3f304102dbdf2c9c49fc9b28259d8de41daa0fc29f3b437f46a90","id":"ev-a3516f711bd3819e104e7503329d647bbe3f2a70553513aef6f59f0bba6df695","recordSha256":"7a1bf8371d35743dd3300d0be34867383159535ceb31788320a04eb2ba315230","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":7220},"recordId":"SYN-PUB-07220"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0633","excerptSha256":"21ff234b9a3afd93e725b87c798514eeed3ef3801a425051d9ff69744522f498","id":"ev-98d252f9384cd20095183d1502bbf10e8c4bf4feaeb7b0cb39562e1151375588","recordSha256":"7a1bf8371d35743dd3300d0be34867383159535ceb31788320a04eb2ba315230","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":7220},"recordId":"SYN-PUB-07220"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"dynamic pricing\",\"revenue management\",\"make-to-order job shop\",\"capacity uncertainty\",\"queueing theory\",\"price-sensitive demand\",\"workload control\",\"demand elasticity\",\"stochastic scheduling\",\"lead-time quotation\"]","excerptSha256":"199e2590ffe022edd3a512121b360d4e0a6d075a7867d375f0014184d3089864","id":"ev-0887876dd29d92e8a81e0ceb66e3be1a656e64664465b296f6b469dda8b64a24","recordSha256":"6e480dcbae799774ff3871830061c4b18a3f8e0b92f665aba491194b5ccdb4f2","source":{"derivedRecordId":false,"fieldPointer":"/keywords","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/632"},"recordId":"R0633"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0633","excerptSha256":"21ff234b9a3afd93e725b87c798514eeed3ef3801a425051d9ff69744522f498","id":"ev-a71dfb37487b8a0947a6b40d0b6bb4333260febc7ba3a9c620daf29dd926f057","recordSha256":"6e480dcbae799774ff3871830061c4b18a3f8e0b92f665aba491194b5ccdb4f2","source":{"derivedRecordId":false,"fieldPointer":"/id","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/632"},"recordId":"R0633"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"When a job shop produces several product families simultaneously, interactions between queues complicate capacity and pricing analysis. This paper applies decomposition-based queueing network approximations (specifically the parametric decomposition method) to a three-product job shop, where each product's arrival rate is a decreasing linear function of its posted price. A profit-maximising price vector is found by gradient search over the resulting approximate steady-state performance functions. The model is validated against simulation on test instances constructed to match characteristics of a Novi Sad sheet-metal fabrication plant. Approximation errors in throughput and mean waiting time are below 8% across a wide range of utilisation levels. The optimal price vector under the approximate model achieves within 3% of the simulation-optimal revenue. A key finding is that ignoring the queueing interaction between product classes overstates optimal prices by 6–12% at high utilisation, with implications for firms quoting jobs without workload visibility across product lines.","excerptSha256":"ee3949e35cae4331679a820de95b3bc5a83d53d954256f76dd170d39f26e3016","id":"ev-e3a431bb416a785834dff0eea279df5b9b3cdc621b7bc35aa833a67e033074de","recordSha256":"cdda1a8b95932a1b8852d851d89be4d562d1a6c277a8024ca1604c2c9b435457","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":7224},"recordId":"SYN-PUB-07224"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0633","excerptSha256":"21ff234b9a3afd93e725b87c798514eeed3ef3801a425051d9ff69744522f498","id":"ev-3b24869aad0e9da9799c6e75852d19999282384d7cb4cc8e657fcbffd205db8f","recordSha256":"cdda1a8b95932a1b8852d851d89be4d562d1a6c277a8024ca1604c2c9b435457","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":7224},"recordId":"SYN-PUB-07224"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Some manufacturers maintain finished-goods buffers for standard items while also accepting custom orders on a make-to-order basis, creating a hybrid production environment where buffer-stock and pricing policies interact. This paper formulates a Markov decision process for a firm choosing both the reorder point of its standard item and the price it quotes for custom orders, given machine capacity shared between both streams. The state space is characterised by buffer inventory level and current custom-order queue length. Value iteration is used to solve instances with up to 40 inventory levels and 20 queue states. Results show that ignoring the interaction between buffer policy and custom-order pricing leads to a 9% average profit loss. The optimal policy exhibits intuitive threshold structure: custom-order prices rise as both buffer stock falls and custom-order queue grows. A sensitivity study over breakdown rates and demand variability parameters is included.","excerptSha256":"07bd6b017893f2c43c8c38a751c68f6ca0330c38ac2c906916e6d48ef3cd7cd3","id":"ev-07087aa1768ec9fa40ec4198445b1c7eaa24df514d081891e658537cbe923221","recordSha256":"14ff3c91cbf7439828afecfd1f81c46725a29a08b4b95935760fdd512e29f4ec","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":7227},"recordId":"SYN-PUB-07227"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0633","excerptSha256":"21ff234b9a3afd93e725b87c798514eeed3ef3801a425051d9ff69744522f498","id":"ev-aa6546ea7955c73090ed8e5780d3e866be2e93b9ac6568c3c1da6dee1e88c3ba","recordSha256":"14ff3c91cbf7439828afecfd1f81c46725a29a08b4b95935760fdd512e29f4ec","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":7227},"recordId":"SYN-PUB-07227"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Standard Wright's law assumes that cumulative average unit cost declines indefinitely as a power function of cumulative output, but in high-mix low-volume contract manufacturing the workforce frequently switches between product types, limiting the depth of learning achievable before a product is discontinued. This paper tests whether a plateau-modified learning curve, which introduces an asymptotic cost floor parameter, fits ramp-up data better than the standard two-parameter model in such environments. A dataset of 22 product ramp-ups from a medical device sub-assembly contractor was used. Model comparison was performed using corrected Akaike information criterion. The plateau model produced a lower AICc in 16 of 22 cases, with the improvement being statistically meaningful in 11 cases based on a likelihood-ratio test at the 0.05 level. The estimated plateau was reached on average at 340 cumulative units, which in this factory corresponded to roughly eight to twelve weeks of production. The paper concludes that the plateau model is preferable when typical production runs are short, and provides a simple heuristic for practitioners to choose between the two formulations based on expected run length.","excerptSha256":"54cf05ffa1ed75633de5048aa07e3f7d7ccdeb8dabc99f167e56d3b71fc4c9c1","id":"ev-0b1c73ccd5dece86293b104d3d9d8580a0f67f23ecb409d88be753d8277405b2","recordSha256":"033367a4fbf18f669c5782dfaa9b29f739d53d8cbe13632f38f6a22ceb39bbc1","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":7067},"recordId":"SYN-PUB-07067"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0633","excerptSha256":"21ff234b9a3afd93e725b87c798514eeed3ef3801a425051d9ff69744522f498","id":"ev-ee1576157cd057f6c3283442fd11a5a0d385691d94aadc8e5503a1987ecb51a2","recordSha256":"033367a4fbf18f669c5782dfaa9b29f739d53d8cbe13632f38f6a22ceb39bbc1","source":{"derivedRecordId":false,"fieldPointer":"/authors/1/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":7067},"recordId":"SYN-PUB-07067"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"Industrial Engineering and Operations Management\",\"Revenue Management and Pricing\",\"Queueing Theory and Stochastic Modelling\",\"Manufacturing Systems\"]","excerptSha256":"44c7a358ea460abf1c7ef2f39dfbe009708524b073750bc6afb25c7e633a7d0e","id":"ev-ed5e4878f5088766cad0254a88e5a77937d7bc7b3f97a1a04da2f65d84d5fcfc","recordSha256":"6e480dcbae799774ff3871830061c4b18a3f8e0b92f665aba491194b5ccdb4f2","source":{"derivedRecordId":false,"fieldPointer":"/fields","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/632"},"recordId":"R0633"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Workload visibility as a prerequisite for dynamic pricing: an assessment of ERP and MES readiness in Serbian manufacturing SMEs","excerptSha256":"d138dd59f4e3090013796926e3a706c7a3f2c7641bce47ae75f639fde0553a0f","id":"ev-60d229d88be51dbebb94c8040f45e2ee3a4fd61825c20ac08619d1473c7b6012","recordSha256":"f5c85a83027413941634dee980ad68ce5e741ef28ce9b13a940659d66e185be3","source":{"derivedRecordId":false,"fieldPointer":"/title","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":7229},"recordId":"SYN-PUB-07229"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0633","excerptSha256":"21ff234b9a3afd93e725b87c798514eeed3ef3801a425051d9ff69744522f498","id":"ev-e62fae6c4576da9b24cf766885ec485b8f50d1554a096a26e2968351c6cb73c9","recordSha256":"f5c85a83027413941634dee980ad68ce5e741ef28ce9b13a940659d66e185be3","source":{"derivedRecordId":false,"fieldPointer":"/authors/1/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":7229},"recordId":"SYN-PUB-07229"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Optimal pricing functions derived from queueing or simulation models are computationally expensive to evaluate in real time, limiting their use in shop-floor execution systems. This paper investigates whether gradient-boosted regression trees trained on offline simulation data can serve as fast surrogates for a six-machine, ten-job-class job shop pricing model. A training dataset of 120,000 state-action-reward triples is generated from AnyLogic simulations. The trained surrogate predicts the near-optimal price for a new arriving job within 2 milliseconds on standard hardware, compared with 40 seconds for direct simulation lookahead. In a rolling-horizon production test, the surrogate policy achieves 97.4% of the revenue attained by the full simulation-based policy. Feature importance analysis identifies current aggregate utilisation and job processing time as the two most influential inputs, with job class contributing less than expected. The paper discusses data requirements and retraining frequency for production deployment.","excerptSha256":"58d3adbf0c26174891c97f4822a1e520728177d62932ef7068b3437230851625","id":"ev-0f3fda55a094e91df75a7b4334e5966a80ccec37602c098577673d8160bda043","recordSha256":"c4b76786536e298ac3223ccccf10781f20b53a0a0d5c8bf9ae9451249ba54637","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1390},"recordId":"SYN-PUB-01390"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0633","excerptSha256":"21ff234b9a3afd93e725b87c798514eeed3ef3801a425051d9ff69744522f498","id":"ev-298e4fbd036a197b22e9be53db7dcf9d2fd140685da2956ffbba50e54077b89f","recordSha256":"c4b76786536e298ac3223ccccf10781f20b53a0a0d5c8bf9ae9451249ba54637","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1390},"recordId":"SYN-PUB-01390"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"The project develops and validates a suite of dynamic pricing models for small and medium-sized make-to-order job shops in Serbia, combining queueing-theoretic workload models with empirically estimated demand elasticity functions. Phase one collected order and capacity data from twelve partner firms and ran stated-preference surveys with their customers. Phase two developed and tested Bayesian learning pricing rules and lightweight surrogate models suitable for deployment in standard ERP environments. Phase three piloted pricing tools at four firms over six months and evaluated revenue and customer satisfaction outcomes. The project produced eight journal papers and one open-source pricing toolkit. Partners include industrial SMEs in the metalworking sector and one software firm supporting ERP customisation.","excerptSha256":"b3ea07ad6dcfbc8b19bd942481fa70e62e781ab01abf3a116a8cc3d47f425758","id":"ev-6154c1db3a6fd6ae32f2bb6a85f8d56906278a6031a3efcaaacc26d1aebfc525","recordSha256":"57bc7e407b8ba506954164897c4f3fa504073d675311baa1d20e878822eeea99","source":{"derivedRecordId":false,"fieldPointer":"/summary","file":"data/projects.json","locator":{"format":"json","jsonPointer":"/projects/854"},"recordId":"SYN-PRJ-0855"},"sourceFileSha256":"847cb13091f91ed9fe57f6aea0088cd1e66ed0263c86a5090f169a2b30ab3a1a"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0633","excerptSha256":"21ff234b9a3afd93e725b87c798514eeed3ef3801a425051d9ff69744522f498","id":"ev-aaf29f7cb3a3891e8f707a211b528798629a4dd422a7be3b923b1c155ca0ce38","recordSha256":"57bc7e407b8ba506954164897c4f3fa504073d675311baa1d20e878822eeea99","source":{"derivedRecordId":false,"fieldPointer":"/participants/0/researcher_id","file":"data/projects.json","locator":{"format":"json","jsonPointer":"/projects/854"},"recordId":"SYN-PRJ-0855"},"sourceFileSha256":"847cb13091f91ed9fe57f6aea0088cd1e66ed0263c86a5090f169a2b30ab3a1a"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"A cross-border cooperation project between the University of Novi Sad and the University of Miskolc aimed at transferring advanced production management methods to manufacturing SMEs in the Vojvodina and Northern Hungary regions. Hadžić led the work package on pricing and demand management, coordinating survey activities across both countries and delivering a two-day training programme for production managers on workload-based pricing. The project produced a bilingual Serbian–Hungarian practical guide on revenue management for job shops, disseminated to 85 firms through regional chambers of commerce. The project also established a shared dataset of anonymised order records for ongoing research use.","excerptSha256":"cc090a5532e78158390f9474aad60342220ef692ac8c750c6917cb86e9931509","id":"ev-e2938f5bd1f24125c4388d15407a30397202a26b16471b40b63aefe3532b7f3b","recordSha256":"04e171732d4404ab622b273098fd63946e0a0a05f70c44884943e2aa637fdab6","source":{"derivedRecordId":false,"fieldPointer":"/summary","file":"data/projects.json","locator":{"format":"json","jsonPointer":"/projects/855"},"recordId":"SYN-PRJ-0856"},"sourceFileSha256":"847cb13091f91ed9fe57f6aea0088cd1e66ed0263c86a5090f169a2b30ab3a1a"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0633","excerptSha256":"21ff234b9a3afd93e725b87c798514eeed3ef3801a425051d9ff69744522f498","id":"ev-ddca91dce2605edda549935ad08b5bc8bfaffd211d995e6f098979ea22575052","recordSha256":"04e171732d4404ab622b273098fd63946e0a0a05f70c44884943e2aa637fdab6","source":{"derivedRecordId":false,"fieldPointer":"/participants/0/researcher_id","file":"data/projects.json","locator":{"format":"json","jsonPointer":"/projects/855"},"recordId":"SYN-PRJ-0856"},"sourceFileSha256":"847cb13091f91ed9fe57f6aea0088cd1e66ed0263c86a5090f169a2b30ab3a1a"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Dragan Hadžić received his Dipl.Ing. in Industrial Engineering from the Faculty of Technical Sciences, University of Novi Sad, in 2003, and his PhD from the same institution in 2010, with a dissertation on workload-dependent pricing in make-to-order manufacturing systems. He joined the Department of Industrial Engineering and Management as a teaching assistant in 2004 and was promoted to Associate Professor in 2018. His research group focuses on dynamic pricing mechanisms, revenue management, and queueing-theoretic capacity models for job-shop and custom-manufacturing environments. He teaches courses in production planning, operations research, and pricing strategy at both undergraduate and graduate levels. The group uses simulation software (Arena, AnyLogic) and statistical computing environments (R, Python) for model estimation and validation, and has access to the Faculty's industrial simulation laboratory.","excerptSha256":"a101fb9cee867ea09934a6c6e679d746c3b5c3b01b5571f842a3cd882172c495","id":"ev-1a4810f632942b206228a9770eece08c8bec44b6bc4755ba640c419f1a647db1","recordSha256":"6e480dcbae799774ff3871830061c4b18a3f8e0b92f665aba491194b5ccdb4f2","source":{"derivedRecordId":false,"fieldPointer":"/bio","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/632"},"recordId":"R0633"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Hadžić is particularly interested in partnerships with NTP Novi Sad companies that deal with custom or project-based manufacturing, software for production management, or supply chain analytics. He prefers collaborations with a clear data-sharing component and is open to co-developing IP if a project has commercial potential.","excerptSha256":"66ef74affe841a9c0496c09d67a0175ae0bec75fbd996def2f54c1d62aab76a4","id":"ev-586db0b56c1ea1bf559c3bb7aeff96dc4bee42e980618854aec55a11697db309","recordSha256":"6e480dcbae799774ff3871830061c4b18a3f8e0b92f665aba491194b5ccdb4f2","source":{"derivedRecordId":false,"fieldPointer":"/notes","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/632"},"recordId":"R0633"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"joint projects with manufacturing SMEs on pricing tool development and deployment\",\"contract research on demand elasticity estimation from company order data\",\"industrial PhD supervision for candidates from manufacturing companies\",\"consulting on revenue management and dynamic quoting system design\",\"student internships in partner firms focused on ERP data analysis and pricing\",\"access to industrial simulation laboratory for joint model validation\"]","excerptSha256":"ce7d30be76723b4d5b1143b355f74c284cd65f5316130201a8398add4d0b5c65","id":"ev-de20ee7e4bf496a1fa1f79d0ea36d31d02c0805c517323760d1705767818c285","recordSha256":"6e480dcbae799774ff3871830061c4b18a3f8e0b92f665aba491194b5ccdb4f2","source":{"derivedRecordId":false,"fieldPointer":"/open_to","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/632"},"recordId":"R0633"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Hadžić has served as a part-time consultant to two Vojvodina metalworking firms on production planning and quoting process redesign, and collaborated with a Novi Sad ERP software provider on integrating workload dashboards into their SME product.","excerptSha256":"47dab48d1a22be0136a866cb4f171f1670cfa1c613d5870e6aa76406d2549223","id":"ev-e488dc332b1ac4ec9a74cbff59f75e8b8ab1ec554fcc569a71677f36dd100767","recordSha256":"6e480dcbae799774ff3871830061c4b18a3f8e0b92f665aba491194b5ccdb4f2","source":{"derivedRecordId":false,"fieldPointer":"/industry_experience","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/632"},"recordId":"R0633"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Revenue management systems can generate rapid within-day rate changes that expose consumers to stark price anchoring effects. This paper reports two controlled online experiments (n = 410 and n = 385) in which Serbian and Hungarian participants evaluated hotel offers under manipulated sequences of displayed price histories. Experiment 1 tests whether an upward reference price trajectory increases willingness to book relative to a flat or declining sequence, holding the current price constant. Experiment 2 introduces a stated 'algorithm-based pricing' explanation and measures its effect on perceived fairness. Results confirm a significant reference price anchoring effect (Cohen's d = 0.41) but show that algorithm attribution reduces perceived fairness scores by 0.6 points on a 7-point scale. No significant moderation by hotel-booking experience is found. The study has implications for how revenue management systems should communicate price rationale to end consumers.","excerptSha256":"69734752d7e1f5db8937fe4873521b9a23b1f6ddacb1d36b72adc7d6e3b13d0c","id":"ev-10335a0a7182a7a7fa018ba75f9dc8d776f89fb098840dcf293a148f7f2438e6","recordSha256":"39ad4038e35a79939c9b800d7fff19820bf3525dc77cc1d0e8fc17947a98f33f","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8291},"recordId":"SYN-PUB-08291"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0751","excerptSha256":"3af26842f8012270bda3d4578cfbaff34694064085c99fe02463d1d5f5d57979","id":"ev-1fa2b3538e6c9caf914d38ad936c220accbfc174c02eb432cb6a66551db4b29a","recordSha256":"39ad4038e35a79939c9b800d7fff19820bf3525dc77cc1d0e8fc17947a98f33f","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8291},"recordId":"SYN-PUB-08291"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Personalised pricing—setting individual rates based on inferred consumer willingness to pay—offers revenue managers potentially large gains but conflicts with European data protection rules under the GDPR. This paper develops a formal model of constrained personalised pricing in which the hotel revenue manager maximises expected revenue subject to a data minimisation constraint that limits the number of consumer attributes usable in rate discrimination. The model is parameterised using conjoint survey data from 502 hotel guests in four countries (Serbia, Hungary, Romania, Austria) and solved numerically. Results show that even a two-attribute rate model (trip purpose, booking channel) recovers 61% of the theoretical full-information revenue gain, while satisfying a plausible interpretation of GDPR data minimisation. The study offers a practical framework that legal and revenue management departments can use jointly to design compliant personalisation strategies.","excerptSha256":"da3cd990c82bd5f45a3c11c2c2c1ff1c8d0cf27ac838b3b973b5e9d5c66cbd93","id":"ev-7c217a306ef4ac445c5a178d9eb3ad6b88bff0a60d2518af76d5336e28b4c6de","recordSha256":"9b15896c0229bcef7bdf4143fc4073888ec13de77b7571bc4cee0586462f4cca","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8293},"recordId":"SYN-PUB-08293"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0751","excerptSha256":"3af26842f8012270bda3d4578cfbaff34694064085c99fe02463d1d5f5d57979","id":"ev-51c28d37e24c46b3b982d216e91b8527e612e91cbaf0c05b3bd6491b5f7aa25b","recordSha256":"9b15896c0229bcef7bdf4143fc4073888ec13de77b7571bc4cee0586462f4cca","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8293},"recordId":"SYN-PUB-08293"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Automated competitive rate monitoring has become a standard component of hotel revenue management, yet the quality of scraped data and the ethical limits of scraping practice receive limited systematic attention in the hospitality literature. This paper conducts a controlled accuracy audit in which rates retrieved by a purpose-built scraping prototype are compared to rates confirmed through direct booking for 48 hotels in Novi Sad, Subotica, and Szeged over a 90-day window totalling 12,960 rate observations. Scraping latency—the delay between a price change and its detection—averages 4.2 hours across the sample, with tails exceeding 18 hours for properties using client-side rendering. Data accuracy is 94.7% at the headline rate level but falls to 81.3% when tax-inclusive rates are required. The paper also reviews applicable terms-of-service restrictions and Serbian and EU legal precedents relevant to automated data collection, providing a compliance checklist for revenue management practitioners.","excerptSha256":"87314eeb81332b6d0e745969e372a1929938e875c10f843e6a130e426d582a21","id":"ev-695d4793204e2ea6616a668670582e7c193b2edd9c15d08fd5b1b04e8519a20f","recordSha256":"7822f97425de0740125ac68ead32096b3701ff0a16efb7c30f640a40059d9c35","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8294},"recordId":"SYN-PUB-08294"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0751","excerptSha256":"3af26842f8012270bda3d4578cfbaff34694064085c99fe02463d1d5f5d57979","id":"ev-8750aacb491f0ca76fdc8005a9c564e7c89bae26ca0b33f7f3bdda895be76ec2","recordSha256":"7822f97425de0740125ac68ead32096b3701ff0a16efb7c30f640a40059d9c35","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8294},"recordId":"SYN-PUB-08294"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Igor Krstić completed his undergraduate and master's studies at the Faculty of Economics in Subotica, University of Novi Sad, and received his PhD in 2009 from the same institution with a dissertation on demand forecasting under price uncertainty in Adriatic coastal tourism. He joined the Department of Management as an assistant professor in 2010 and was promoted to full professor in 2021. His research group focuses on dynamic pricing models, revenue management algorithms, and the behavioural responses of consumers to price volatility in hotels, short-term rental platforms, and airline ancillary services. He teaches courses in strategic management, tourism economics, and quantitative methods for business. The group regularly uses statistical software environments and publicly available booking-platform datasets; no dedicated wet-lab or instrumentation is maintained.","excerptSha256":"2c554218661c572eeaaa6eb6b292b6843e71b84fcad6eec592ffba4b1a221069","id":"ev-980a05c82dc631df58266f67b38d3ec1602e8a09a09bfa998468c90f7bdb819e","recordSha256":"68878d3c686a6f4c79725a9393453578cb6b3338e749a4ddaaddd5263170628c","source":{"derivedRecordId":false,"fieldPointer":"/bio","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/750"},"recordId":"R0751"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0751","excerptSha256":"3af26842f8012270bda3d4578cfbaff34694064085c99fe02463d1d5f5d57979","id":"ev-0858db3454982cdbd376973d8c49341d1fe2604c745a117b154558c9e40b3615","recordSha256":"68878d3c686a6f4c79725a9393453578cb6b3338e749a4ddaaddd5263170628c","source":{"derivedRecordId":false,"fieldPointer":"/id","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/750"},"recordId":"R0751"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"dynamic pricing\",\"revenue management\",\"demand uncertainty\",\"asymmetric information\",\"hospitality economics\",\"price forecasting\",\"yield management\",\"tourism markets\",\"price elasticity\",\"overbooking\"]","excerptSha256":"e4a558c36483a928b0a6c2066b7071767749ea6351de15c97a3e433adb1559dd","id":"ev-0bea5a1ae73bad8c92940af53bbd52880a09cc1276131479f58a9579a01c08db","recordSha256":"68878d3c686a6f4c79725a9393453578cb6b3338e749a4ddaaddd5263170628c","source":{"derivedRecordId":false,"fieldPointer":"/keywords","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/750"},"recordId":"R0751"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"The COVID-19 pandemic produced an unprecedented and heterogeneous demand shock in urban hotel markets. Using daily administrative revenue data from 31 hotels in Belgrade, Novi Sad, Kragujevac, and Niš for the period January 2019 to December 2022, this paper estimates city-specific demand recovery curves and evaluates whether hotels that adopted aggressive repricing during the trough phase recovered faster than those that maintained pre-pandemic rate levels. A synthetic control method is applied to construct counterfactual recovery paths for each property. Results show that hotels that cut average daily rates by more than 20% during the trough recovered to 90% of pre-pandemic RevPAR roughly 4.2 months earlier on average, but at a cost of sustained brand dilution measurable in online review scores 18 months later. The trade-off between short-run occupancy gains and longer-run reputation effects is quantified and discussed.","excerptSha256":"15a3237720cc283f2be918f73d0a21d709b37a97d3209135258fd21897b42283","id":"ev-4d55ff147dd15049a8b5a09cae6650c10842159ae717d985ab99846fe8a7aa91","recordSha256":"6c0b2bada7d9faa6e769025c8f70f72b35dc9ac0c9d8be308415c30a57c34169","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1650},"recordId":"SYN-PUB-01650"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0751","excerptSha256":"3af26842f8012270bda3d4578cfbaff34694064085c99fe02463d1d5f5d57979","id":"ev-2f69148a44b06a3645a0d4fdad10dd9822de647a3244dbcee89168566988d581","recordSha256":"6c0b2bada7d9faa6e769025c8f70f72b35dc9ac0c9d8be308415c30a57c34169","source":{"derivedRecordId":false,"fieldPointer":"/authors/3/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1650},"recordId":"SYN-PUB-01650"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Online travel agencies have made price dispersion in tourism accommodation highly visible to consumers, yet its structural determinants remain understudied in South-East European contexts. This study draws on a scraped dataset of 4,200 daily rate observations from a single mid-size online travel agency covering 112 Dalmatian coastal hotels during the summer seasons of 2008 and 2009. Quantile regression estimates reveal that price dispersion at the upper tail is driven primarily by star-rating heterogeneity and last-minute inventory clearance, while dispersion at the lower tail reflects competitive pressure from unlicensed accommodation. A price-concentration index constructed for 14 micro-destinations shows significant variation even within the same coastal stretch. The findings suggest that revenue managers in the region treat promotional discounting and peak-period premiums as largely independent decisions, which limits overall yield performance.","excerptSha256":"6b763ad0a45bff98afe2bd558a3f8136399014a904827eede2f6d8fb57962001","id":"ev-d08536a3254d12c610c8482703efa1caa1bc614021c47fdc12fc35d304d392f0","recordSha256":"9ce271b0d766e83a20e0f33b6990e0950561673d3074290f4cc2574e5277a45f","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8285},"recordId":"SYN-PUB-08285"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0751","excerptSha256":"3af26842f8012270bda3d4578cfbaff34694064085c99fe02463d1d5f5d57979","id":"ev-b47321e03f5d3c0af7bdaf87ef7ef3a9eddf51cd10103138373c515a2bcb647b","recordSha256":"9ce271b0d766e83a20e0f33b6990e0950561673d3074290f4cc2574e5277a45f","source":{"derivedRecordId":false,"fieldPointer":"/authors/1/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8285},"recordId":"SYN-PUB-08285"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Reference Price Effects and Perceived Fairness Under Algorithmic Hotel Pricing: An Experimental Study","excerptSha256":"6b1990214233c48357e47a48659d3c9209778942f2fe809118d5f529b5c1bf63","id":"ev-61c27b7add0c41a094f74677ff3badae6625a4c9ced5e2a08e7ab56202958afa","recordSha256":"39ad4038e35a79939c9b800d7fff19820bf3525dc77cc1d0e8fc17947a98f33f","source":{"derivedRecordId":false,"fieldPointer":"/title","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8291},"recordId":"SYN-PUB-08291"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Hotels in Serbian cities began introducing early check-in upselling programmes around 2011–2013, yet evidence on their net revenue contribution is sparse. Exploiting a natural experiment created when a Novi Sad hotel chain introduced a structured upselling script for front-desk staff in mid-2013 while two comparable properties delayed implementation, this study applies a difference-in-differences estimator to daily revenue per available room (RevPAR) data for a 28-month period. The treated properties show a statistically significant increase in ancillary revenue of approximately 4.2 EUR per occupied room, with no detectable displacement of room revenue. Heterogeneity analysis shows the effect is concentrated in leisure segments and disappears for corporate accounts managed under contracted rates. The study illustrates how relatively low-cost operational changes can complement algorithmic pricing tools.","excerptSha256":"70b8b022b84c794a1cc60615795d89546cf71a7087eabc83eefeda44aeecf4ce","id":"ev-dcd801777697e78e36226801ee7f756e5a13643180b713fc5fdbff330d3c9a55","recordSha256":"2abed99d377d72dcabc7d6ed3817937beb49a5b790b94da7314b8a71e1c65d8d","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8287},"recordId":"SYN-PUB-08287"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0751","excerptSha256":"3af26842f8012270bda3d4578cfbaff34694064085c99fe02463d1d5f5d57979","id":"ev-9d7b4510d7a0db279b08edafbdcddc44c717d1f17ae9fb41683eb3d53ae84b9f","recordSha256":"2abed99d377d72dcabc7d6ed3817937beb49a5b790b94da7314b8a71e1c65d8d","source":{"derivedRecordId":false,"fieldPointer":"/authors/0/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8287},"recordId":"SYN-PUB-08287"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Platform Competition and Rate Parity Erosion in Serbian Short-Term Rental Markets, 2014–2018","excerptSha256":"7637f01889acdd87b25aa1e1f976af94cfc1de750b38be5c9f4e8bb4e8fb59a1","id":"ev-0f1af8e3bce2ba9ef6b466a7b10ac961ceb84cff404c9af0d6b2c7bcb8ae531c","recordSha256":"5d4fd781c3e7626c48d22fffdf9c41136cebfcc9ec08d7d1c46a7d8c7d2b8fb6","source":{"derivedRecordId":false,"fieldPointer":"/title","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1649},"recordId":"SYN-PUB-01649"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0751","excerptSha256":"3af26842f8012270bda3d4578cfbaff34694064085c99fe02463d1d5f5d57979","id":"ev-71ab29f255d7f7759b74177242923b5344882d54f29d950a4dffa8e7d9e9904d","recordSha256":"5d4fd781c3e7626c48d22fffdf9c41136cebfcc9ec08d7d1c46a7d8c7d2b8fb6","source":{"derivedRecordId":false,"fieldPointer":"/authors/1/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":1649},"recordId":"SYN-PUB-01649"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"The project develops and validates a revenue management decision-support toolkit tailored to independently owned accommodation businesses in the Pannonian region. Core activities include assembly of a multi-year daily rate and occupancy panel from partner hotels and rental properties, estimation of asymmetric demand models that distinguish high-side and low-side forecast errors, and design of rule-based pricing triggers suitable for operators without dedicated revenue management staff. Fieldwork involves structured interviews with 80 accommodation managers and a randomised advisory intervention with 22 willing participants. Deliverables include an open-access dataset, a calibrated forecasting module, and a practitioner handbook distributed through the Tourism Organisation of Vojvodina.","excerptSha256":"1e452a988ad25e9708aeb20339c820b1591e200ec6e2d9202e8b2635fad8f3d6","id":"ev-311277601e4a0e8e99883fa9a8ddf436b9fdee32a85c492033a57b370913be39","recordSha256":"efdad8ccb2df6a8af32676fd51368818ce4d5933f44d21f142e597c4d232732b","source":{"derivedRecordId":false,"fieldPointer":"/summary","file":"data/projects.json","locator":{"format":"json","jsonPointer":"/projects/1619"},"recordId":"SYN-PRJ-1620"},"sourceFileSha256":"847cb13091f91ed9fe57f6aea0088cd1e66ed0263c86a5090f169a2b30ab3a1a"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0751","excerptSha256":"3af26842f8012270bda3d4578cfbaff34694064085c99fe02463d1d5f5d57979","id":"ev-27718225bb452042820a18d1bed15d57a3ade7420f2f1066100c7cbb8ce4f0af","recordSha256":"efdad8ccb2df6a8af32676fd51368818ce4d5933f44d21f142e597c4d232732b","source":{"derivedRecordId":false,"fieldPointer":"/participants/0/researcher_id","file":"data/projects.json","locator":{"format":"json","jsonPointer":"/projects/1619"},"recordId":"SYN-PRJ-1620"},"sourceFileSha256":"847cb13091f91ed9fe57f6aea0088cd1e66ed0263c86a5090f169a2b30ab3a1a"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Agritourism properties in Vojvodina operate under capacity constraints far tighter than urban hotels and serve communities where visible price discrimination may carry social costs. This book chapter synthesises 14 semi-structured interviews with agritourism operators and a discrete-choice experiment with 380 potential guests to assess the acceptability and revenue potential of demand-responsive pricing in this segment. Interview analysis identifies three adoption barriers: lack of forecasting skills, fear of negative word-of-mouth, and perceived incompatibility with the 'authentic rural' brand. The discrete-choice experiment shows that guests tolerate price premiums of up to 18% during peak agritourism weekends but penalise perceived arbitrariness by switching to competitors at premiums above 25%. The chapter concludes with a practical tiered pricing framework calibrated to these tolerance thresholds and compatible with simple spreadsheet management.","excerptSha256":"222f4011174407b3b589ad7a287e15b81871b81ff857114e7b5fd91835dbf1fe","id":"ev-6be61bd6b9d68e3ebc593f78c7d25b16f5369e8c3ad6c53e72181d28a7e48dc4","recordSha256":"6fc512c2d246ce23363e6e7412287478af11b84ec135f622eaf405cc1999aa6d","source":{"derivedRecordId":false,"fieldPointer":"/abstract","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8290},"recordId":"SYN-PUB-08290"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"R0751","excerptSha256":"3af26842f8012270bda3d4578cfbaff34694064085c99fe02463d1d5f5d57979","id":"ev-4636115fea3c5991536c9bb78f2be6b4d6d293223c9e22379ae2dbf7fa6992d3","recordSha256":"6fc512c2d246ce23363e6e7412287478af11b84ec135f622eaf405cc1999aa6d","source":{"derivedRecordId":false,"fieldPointer":"/authors/2/researcher_id","file":"data/publications.jsonl","locator":{"format":"jsonl","line1Based":8290},"recordId":"SYN-PUB-08290"},"sourceFileSha256":"25d8be510899a943cbed484ae7b7587ca81d538bbe736eff065e928b3b3c0019"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Krstić is particularly interested in collaboration with companies building property management or revenue management software and with data providers holding large accommodation booking datasets. He prefers projects with a clear applied component that can feed back into teaching cases. He is not available for short consulting engagements of less than three months.","excerptSha256":"25ec764fd9821eda5b676210f7e4674afda18afcd4d16226dde894b3bd8901c1","id":"ev-ecc67f71ee98108aefe6920a6811a10cbdce05b81fe0aa788868f6e1453c26ed","recordSha256":"68878d3c686a6f4c79725a9393453578cb6b3338e749a4ddaaddd5263170628c","source":{"derivedRecordId":false,"fieldPointer":"/notes","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/750"},"recordId":"R0751"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"[\"joint projects\",\"contract research\",\"consulting\",\"student internships\",\"industrial PhD supervision\"]","excerptSha256":"d29497a9842abdbaa513213cd745db1c53b73ff4e4d17a196497774f24877241","id":"ev-93b4ce53b8f3fb18dab862e3f25644d2120c762f98e4305a76597f3a4e2cdabb","recordSha256":"68878d3c686a6f4c79725a9393453578cb6b3338e749a4ddaaddd5263170628c","source":{"derivedRecordId":false,"fieldPointer":"/open_to","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/750"},"recordId":"R0751"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"},{"canonicalizationVersion":"node-stable-json-v1","datasetCommit":"4fb99b3749adf0d10b5977089ff4c73f4f522240","excerpt":"Krstić has provided pricing strategy consulting to two regional hotel groups in Vojvodina on a contract basis since 2017. He served on the advisory board of a Novi Sad-based hospitality technology startup between 2020 and 2023.","excerptSha256":"e7a49e1268acbf9e5d41efdb54ca0ba62ca2c1ca8f3226b1f41134670856c437","id":"ev-26c605a94d9bdf1a03f13f9a0c22271f4c9f2982c75b3c31ee57fcc3980c62d4","recordSha256":"68878d3c686a6f4c79725a9393453578cb6b3338e749a4ddaaddd5263170628c","source":{"derivedRecordId":false,"fieldPointer":"/industry_experience","file":"data/researchers.json","locator":{"format":"json","jsonPointer":"/researchers/750"},"recordId":"R0751"},"sourceFileSha256":"00de62344d92b6c41c8b67c7e39b1cd8db5553d4050f4c4e92df89ea172dde40"}]