Vector Search

Vector search is the retrieval technique that finds the most semantically similar items to a query by comparing high-dimensional embeddings rather than matching keywords. The query is embedded into the same vector space as the stored documents, and an approximate nearest neighbor (ANN) algorithm returns the top N most similar entries. The result is retrieval that understands meaning: a query for "buying a house" returns documents about real estate, mortgages, and home inspection, even if none of those words appear in the query. The two dominant ANN algorithms in production are HNSW (Hierarchical Navigable Small World) and IVF (Inverted File Index). HNSW trades higher memory for very fast query times and is the default in most modern vector databases. IVF and its variants (IVF-PQ for product quantization) trade some recall for much smaller memory footprints and are useful when you need to store hundreds of millions of vectors on commodity hardware. For AI builders, vector search is one half of a hybrid retrieval strategy. Pure vector search excels at semantic matching but can miss exact-keyword cases (product SKUs, error codes, person names); pure keyword search misses paraphrases and synonyms. The strongest production systems combine both — BM25 plus vector — and rerank the merged results with a cross-encoder. Most modern AI search APIs (Exa is the cleanest example) expose this hybrid retrieval as the default.