DataEngBytes 2023 - PER-T1-11 - Mark Kitson
summary
Argues that a semantic layer is a linked set of business and technical definitions that prevents costly misinterpretation, supports governance, and connects metrics to sources; recommends using SKOS/RDF and knowledge graph tech, linking LLMs to structured knowledge to contain hallucination risk, and keeping a slim metrics layer built from agreed terms. Cites implementations (proof‑of‑concept at a major UK bank, integrations with dbt/Looker, thousands of apps) that revealed hidden dependencies and produced actionable alignment.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 2:11Cost of miscommunication in data
- 4:00Defining the semantic layer
- 6:07Data quality pyramid
- 8:56Illusion of understanding in data
- 11:05Semantic layer enables connections
- 14:14Semantic layer vs data modeling
- 16:24Importance of common language
- 18:51Respecting tribal dialects
- 19:13Right level of definition specificity
- 21:20Aligning definitions in stages
- 23:28Using SKOS and semantic tech
- 25:15Structured knowledge vs LLMs
- 26:39Linking LLMs to structured knowledge
- 27:21Semantic layer in impact sandwich
- 29:47Hiding complexity from users
- 31:34Recap and closing remarks
related talks
21mDataEngBytes 2023 - SYD-T3-04 - Abhinav Goyal
42mDataEngBytes 2023 - SYD-T1-02 - Mikiko Bazeley
27mDataEngBytes 2023 - BRI-T1-07 - Joe Reis
49mSilja Märdla | Data Track | Compass AI & Data Summit 2025
11mStrengthening the data value chain
22mPowering Enterprise Feature Stores with a Universal Semantic Layer
28mthe semantic layer your AI needs
26mDataEngBytes 2023 - MEL-T1-01 - Joe Reis
