Ontologies, Context Graphs & Semantic Layers

Semantic layers answer "What is X?" — metrics-first, SQL abstractions, human BI consumption. Ontologies support inference — formal concepts, explicit relationships, system understanding. One optimizes for measurement; the other for meaning.

Knowing revenue = SUM(order_total) WHERE completed doesn't explain why revenue dropped in Q3. Semantic layers modeled metrics; the metadata was structural, not semantic. Logic stayed in syntax/compute rather than representation.

Gene Ontology (bioinformatics, 20+ years), SNOMED CT (healthcare, 350K+ concepts), Palantir Foundry (operational decision-making). These work at scale because the alternative doesn't.

A subtype of knowledge graph capturing decision reasoning — why actions happened, not just outcomes. Example: VP approves discount exceeding policy; context graph records the reasoning. PKO formalizes this for industrial applications.

dbt's MetricFlow encodes meaning in YAML — calculation models absent relationships, natural language, definitions, and rich context. Metadata was structural not semantic. Ontology engineering and metric definition are fundamentally different disciplines.

LLMs need context and meaning, not dashboards. Knowledge graphs and ontologies provide structured, relationship-rich information that AI systems can reason over — purpose-built for context provision in ways that flat metric definitions cannot match.

Talisman is skeptical: adding inference to a SQL-generation system would require building a KG from scratch. A middle ground may emerge via context-aware semantic layers (Open Semantic Interchange).

Emerging category: "the art of providing all the context for the task to be plausibly solvable by the LLM" — encompassing prompts, memories, ontologies, knowledge graphs, and tool descriptions.

(1) Building an AI analyst requires knowledge representation, not just metrics. (2) Ask "Is this for humans or for AI?" when evaluating semantic layers. (3) Ontologies require domain expertise, knowledge management, and iterative refinement.

For metric lookup, semantic layers suffice. For complex reasoning and inference, they won't. Metrics may become one input to a richer architecture. The winners solve for meaning, not calculation.

PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
SELECT ?type (COUNT(?s) AS ?count)
WHERE { GRAPH <https://linkeddata.uriburner.com/DAV/demos/daas/ontologies-context-graphs-semantic-jessica-talisman-deepseek_v4pro-1.ttl> { ?s rdf:type ?type } }
GROUP BY ?type ORDER BY DESC(?count)

PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX schema: <http://schema.org/>
PREFIX : <https://metadataweekly.substack.com/p/ontologies-context-graphs-and-semantic#>
SELECT ?entity ?name ?type WHERE {
  GRAPH <https://linkeddata.uriburner.com/DAV/demos/daas/ontologies-context-graphs-semantic-jessica-talisman-deepseek_v4pro-1.ttl> {
    { ?entity rdf:type schema:Organization ; schema:name ?name }
    UNION { ?entity rdf:type :OntologyExample ; schema:name ?name }
  }
} ORDER BY ?name

DESCRIBE <https://metadataweekly.substack.com/p/ontologies-context-graphs-and-semantic#analysis>
FROM <https://linkeddata.uriburner.com/DAV/demos/daas/ontologies-context-graphs-semantic-jessica-talisman-deepseek_v4pro-1.ttl>

DESCRIBE uses text/x-html-nice-turtle.