Semantic Search
Search that matches on meaning via embeddings, not exact keywords.
Semantic search finds results by meaning rather than literal keyword overlap: both the query and the documents are turned into embeddings, and the nearest vectors are returned — so “car” can match “automobile” with no shared words.
Worked example: embed every document once, embed the query at search time, and return the top-k by cosine similarity (usually via an ANN index like HNSW); it is the retrieval half of most RAG systems. Gotcha: it can miss exact-term matches (a specific error code, a rare name) that keyword search nails, which is why production search is often hybrid — semantic plus BM25 — with a reranker on top.