Semantic Search
Searching by meaning rather than exact keywords, using embeddings.
Simple explanation
Semantic search finds results that mean the same thing, even if the words are different. 'How do I get a refund?' matches 'What is the return process?' because the embeddings are similar.
Engineering explanation
Semantic search embeds the query and finds nearest neighbours in a vector index. Complements keyword search (BM25) — semantic catches paraphrases, BM25 catches exact identifiers. Production uses hybrid (vector + BM25) for robustness. RAG retrieval is semantic search over chunked documents.
Example
Query: 'cancel subscription' → semantic search finds 'How to end your plan' (different words, same meaning).
When to use it
Search over prose, FAQs, knowledge bases. Anywhere users phrase things differently from how content is written.
Common misconception
Semantic search is NOT better than keyword search — they're complementary. Semantic misses exact IDs; keyword misses paraphrases. Use hybrid.