Semantic Search

Semantic search goes beyond traditional keyword matching to understand the intent and meaning behind a search query. Instead of looking for exact word overlap, semantic search converts both queries and documents into vector embeddings and finds documents whose embeddings are closest to the query embedding in vector space.

This approach solves a fundamental limitation of keyword search: it can match "automobile" with "car" or "myocardial infarction" with "heart attack" because semantically similar concepts are mapped to nearby points in embedding space. Semantic search is powered by embedding models like BERT, text-embedding-3, or BGE that have been trained to capture semantic relationships.

While powerful for meaning-based retrieval, semantic search has limitations -- it can miss exact matches for product IDs, error codes, or rare technical terms where keyword-based methods like BM25 excel. This is why production systems typically combine semantic search with keyword search in hybrid retrieval approaches.