Overview
Two worlds, same word
'Semantic layer' means two adjacent things in industry today:
- The modern-data-stack semantic layer (dbt SL, Cube, LookML) — a YAML/JS/LookML catalogue of metrics and dimensions on top of warehouse marts. Pragmatic, BI-first, tabular.
- The ontology-driven semantic layer (Stardog, Anzo / Cambridge Semantics, TopBraid, EKG vendors) — an OWL/SKOS ontology is the model, and queries are answered by rewriting SPARQL or GraphQL into SQL on the fly.
What ontology-driven adds
- Multi-domain navigation — a question like 'show me all Pharma products whose adverse-event signal correlates with manufacturing batch X' crosses six domains. A flat YAML catalogue collapses under the cardinality; an ontology stays graph-shaped.
- Semantic reasoning —
Drug ⊑ ChemicalCompoundis declared once in OWL; a query for 'all chemical compounds in shipment X' transparently includes drugs without the analyst remembering the subclass relationship. - Enterprise vocabulary — finance, regulatory and operations teams share canonical entities (
Customer,Counterparty,Trade,Position) defined once in the ontology, not five times in five YAML repos.
Where it makes sense
- Pharma R&D, financial-services regulatory, oil & gas operations, government data fabrics — domains with many interlinked specialised vocabularies that already publish ontologies (FIBO, IDMP, OSDU).
- SaaS startups with one product and ten dashboards: overkill — go modern-data-stack.
Cross-link: this track's Ontology Engineering and Semantic Web Foundations tracks teach the OWL/SHACL toolset that the ontology-driven layer uses underneath.