Data substrate

Semantic Search

Search that matches meaning instead of exact words — ask in your own phrasing and find the document that answers, even when it uses different vocabulary.

Operating principle

Production AI is a system: context, tools, permissions, traces, evals, and feedback loops around the model.

What it is

Classic search matches keywords; semantic search matches intent. It embeds the query and the corpus into the same vector space and returns what lands closest in meaning. 'How do I get my money back' finds the refunds policy even though the policy never says 'money back'.

Why it matters

People don't know your internal vocabulary. Semantic search closes the gap between how a question is asked and how the answer was written — which is exactly the gap that makes intranet search frustrating and AI assistants useful.

How it works

Semantic search alone misses exact identifiers — product codes, error strings, names — so production retrieval is hybrid: a semantic leg and a keyword leg (BM25 or Postgres full-text) fused with reciprocal rank fusion, often reranked. Semantic search is one leg of the pipeline, not the whole answer.

Related resources