Semantic Search

Overview

Unlike keyword search (e.g. BM25), semantic search converts both query and documents into embedding vectors and ranks by vector distance. This surfaces semantically related documents even when exact terms differ — useful for internal Q&A, FAQ search, and e-commerce product discovery.

When to use each approach

Exact-match scenarios (part numbers, IDs) still favor keyword search. For natural-language queries with synonyms and paraphrases, semantic search wins. In practice, hybrid search combining BM25 and embedding retrieval achieves the highest precision.