--- schema: 1 type: video title: >- Just-in-Time Ontologies: What Cheap Intelligence Means for Data — Hrishi Olickel, Southbridge AI description: >- A pipeline stripped the minus signs from loan amounts and turned a net $784,000 into $5.3 million of debt, with multiple models agreeing on the wrong result. Hrishi Olickel uses failures like this to explain why transforming data requires more than a universal schema or a confident agent. A customer means something different to billing, support and compliance. Useful structure depends on the source, the storage constraints, the decision being made and the assumptions a model brings to the task. His alternative is a just in time ontology: build the transformation when the question is known, and preserve its context and evidence for reuse. An Ohio campaign finance entity resolution project provides the working example. Olickel describes separating models that design a pipeline, models that perform its individual jobs and models that optimize the resulting system. Definitions, held out cases and disagreement checks matter as much as model selection. Comparing different model families can concentrate review on their disagreements, but shared assumptions can still produce the same wrong answer. The optimization loop needs a stable execution surface, bounded budgets, reusable methods and a defined way to fail. Small pilots establish a baseline; independently checked work segments make revisions easier to inspect and revert. Only then can the pipeline trade among cost, speed and accuracy without losing the evidence behind its answers. Speaker info: - https://github.com/hrishioa - https://www.linkedin.com/in/hrishioa/ Timestamps: 00:00 - Cheap transformation and an Ohio data pipeline 1:44 - Why a universal schema breaks down 4:28 - Building an ontology for the decision 5:34 - Three kinds of data workload 8:30 - Designers, workers and optimizing models 10:34 - Entity resolution and definition checks 13:58 - Model priors and silent data errors 15:41 - Cross model checks and shared mistakes 16:42 - Four requirements for self optimization 18:14 - Bounded pilots and reusable checked segments --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/Fa3hQVIKShw-youtube-fe1d3407674563dc3e28 Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- canonical: >- https://liquidgold.cc/channel/aidotengineer/explore/video/just-in-time-ontologies-what-cheap-intelligence-mea-192f159fe30e markdown: >- https://liquidgold.cc/channel/aidotengineer/video/just-in-time-ontologies-what-cheap-intelligence-mea-192f159fe30e.md llms: null parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-23T22:24:15.954Z' language: null children: [] videoId: Fa3hQVIKShw durationSeconds: 1263 publishedAt: '2026-10-12T16:00:35.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 4 endSeconds: 15 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=4' heldReason: rights - ordinal: 1 text: null startSeconds: 15 endSeconds: 105 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=15' heldReason: rights - ordinal: 2 text: null startSeconds: 105 endSeconds: 116 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=105' heldReason: rights - ordinal: 3 text: null startSeconds: 116 endSeconds: 275 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=116' heldReason: rights - ordinal: 4 text: null startSeconds: 116 endSeconds: 275 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=116' heldReason: rights - ordinal: 5 text: null startSeconds: 275 endSeconds: 285 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=275' heldReason: rights - ordinal: 6 text: null startSeconds: 285 endSeconds: 340 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=285' heldReason: rights - ordinal: 7 text: null startSeconds: 340 endSeconds: 350 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=340' heldReason: rights - ordinal: 8 text: null startSeconds: 350 endSeconds: 520 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=350' heldReason: rights - ordinal: 9 text: null startSeconds: 520 endSeconds: 530 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=520' heldReason: rights - ordinal: 10 text: null startSeconds: 530 endSeconds: 638 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=530' heldReason: rights - ordinal: 11 text: null startSeconds: 530 endSeconds: 638 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=530' heldReason: rights - ordinal: 12 text: null startSeconds: 649 endSeconds: 842 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=649' heldReason: rights - ordinal: 13 text: null startSeconds: 649 endSeconds: 842 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=649' heldReason: rights - ordinal: 14 text: null startSeconds: 842 endSeconds: 852 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=842' heldReason: rights - ordinal: 15 text: null startSeconds: 852 endSeconds: 950 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=852' heldReason: rights - ordinal: 16 text: null startSeconds: 950 endSeconds: 960 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=950' heldReason: rights - ordinal: 17 text: null startSeconds: 960 endSeconds: 1003 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=960' heldReason: rights - ordinal: 18 text: null startSeconds: 1015 endSeconds: 1102 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=1015' heldReason: rights - ordinal: 19 text: null startSeconds: 1102 endSeconds: 1113 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=1102' heldReason: rights - ordinal: 20 text: null startSeconds: 1102 endSeconds: 1113 watch: 'https://liquidgold.cc/watch/Fa3hQVIKShw?t=1102' heldReason: rights --- # Just-in-Time Ontologies: What Cheap Intelligence Means for Data — Hrishi Olickel, Southbridge AI A pipeline stripped the minus signs from loan amounts and turned a net $784,000 into $5.3 million of debt, with multiple models agreeing on the wrong result. Hrishi Olickel uses failures like this to explain why transforming data requires more than a universal schema or a confident agent. A customer means something different to billing, support and compliance. Useful structure depends on the source, the storage constraints, the decision being made and the assumptions a model brings to the task. His alternative is a just in time ontology: build the transformation when the question is known, and preserve its context and evidence for reuse. An Ohio campaign finance entity resolution project provides the working example. Olickel describes separating models that design a pipeline, models that perform its individual jobs and models that optimize the resulting system. Definitions, held out cases and disagreement checks matter as much as model selection. Comparing different model families can concentrate review on their disagreements, but shared assumptions can still produce the same wrong answer. The optimization loop needs a stable execution surface, bounded budgets, reusable methods and a defined way to fail. Small pilots establish a baseline; independently checked work segments make revisions easier to inspect and revert. Only then can the pipeline trade among cost, speed and accuracy without losing the evidence behind its answers. Speaker info: - https://github.com/hrishioa - https://www.linkedin.com/in/hrishioa/ Timestamps: 00:00 - Cheap transformation and an Ohio data pipeline 1:44 - Why a universal schema breaks down 4:28 - Building an ontology for the decision 5:34 - Three kinds of data workload 8:30 - Designers, workers and optimizing models 10:34 - Entity resolution and definition checks 13:58 - Model priors and silent data errors 15:41 - Cross model checks and shared mistakes 16:42 - Four requirements for self optimization 18:14 - Bounded pilots and reusable checked segments --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/Fa3hQVIKShw-youtube-fe1d3407674563dc3e28 Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- Canonical: https://liquidgold.cc/channel/aidotengineer/explore/video/just-in-time-ontologies-what-cheap-intelligence-mea-192f159fe30e · Channel: [AI Engineer](https://liquidgold.cc/channel/aidotengineer.md) > Search chunk text is held until public-source provenance is verified. ## Document chunks --- #### Chunk 0: Document excerpt (0:04 to 1:45) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 1: Document excerpt (1:45 to 4:35) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 2: Document excerpt (4:35 to 5:40) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 3: Document excerpt (5:40 to 8:40) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 4: Document excerpt (8:40 to 10:38) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 5: Document excerpt (10:38 to 14:02) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 6: Document excerpt (14:02 to 15:50) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 7: Document excerpt (15:50 to 16:43) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 8: Document excerpt (16:43 to 18:22) > Held: transcript rights do not permit this chunk's text. --- #### Chunk 9: Document excerpt (18:22 to 21:03) > Held: transcript rights do not permit this chunk's text.