Ontologies, Context Graphs, and Semantic Layers

A semantic layer tells you what your revenue is by normalizing metric definitions. An ontology represents that a customer is a class with specific attributes, placed an order related to products through defined relationships in a market with specific characteristics. Semantic layers are built for analysis (human consumption via BI tools); ontologies are built for reasoning (systems and AI understanding domains to disambiguate, discover context, make inferences).

A context graph is a knowledge graph that captures decision reasoning and operational context, answering "Why was X allowed to happen?" rather than just "What happened?" It includes procedure specifications, execution histories, agent roles and authority, and audit trails.

LLMs need context and meaning, not dashboards. They need to understand what things are, how they relate, and what actions are possible. Semantic layers provide metric definitions optimized for human consumption. Ontologies and knowledge graphs provide concepts, relationships, and constraints that give data meaning, supporting inference and reasoning.

YAML configurations capture SQL table and column representations but are absent of relationships, natural language, definitions, and rich context. They are models of calculations (database entities), not models of the business.

Context engineering is the discipline of designing a system that provides the right information and tools, in the right format, to give an LLM everything it needs to accomplish a task. It includes prompts, memories, few-shot examples, and tool descriptions.

PKO is an ontology developed by Cefriel with Siemens and BOSCH that distinguishes between procedures (abstract specifications) and executions (concrete instances). It organizes knowledge across six areas: procedure specs, action steps, change tracking, execution histories, agent roles, and supporting documentation.

Adding classes, properties, and inference rules to a system designed for SQL generation may require such fundamental architectural changes that you'd essentially be building a knowledge graph from scratch. The gap between metric definitions and formal ontologies may be too wide to bridge.

While the BI industry was perfecting LookML in 2012, Palantir was scaling Foundry with ontologies over semantic layers, building for operational decision making where understanding entity relationships and causal chains was mission-critical. Their architecture was context-first, not metrics-first.