A simple semantic layer a retrieval agent can call

A semantic layer is essential for ensuring an agent answering questions about a company's numbers provides accurate and reliable responses. The layer closes failures by translating business terms into governed concepts, fetching certified metric definitions and running queries on certified tables. This infrastructure is comprised of four key components: a metrics layer, a catalog, an ontology, and a knowledge graph.

At the heart of the semantic layer, a compiled bundle streamlines three critical calls: resolving business terms, fetching certified definitions, and running queries. The first tool maps business terms to governed concepts, providing a unique identifier, label, definition, and associated metric. The second tool retrieves the certified definition, including the formula, unit, version, and query block that dictates how the metric should be computed.

The third tool executes the query on the certified table, returning the requested data while adhering to specified filters and constraints.

By consolidating these functions into a single compiled bundle, the semantic layer achieves higher reliability, performance, and consistency compared to more intricate solutions. Each component is a standalone service with its own definition of a metric and operational footprint. The bundle encapsulates all necessary logic and metadata in-memory, eliminating the need for external tools or APIs during runtime.

The bundle is version-controlled, ensuring that definitions remain intact and up-to-date without the necessity for runtime schema revisions. This streamlined architecture enables the agent to deliver precise and accurate answers to user queries while adhering to defined constraints and guidelines.