Vector search techniques
Algorithms for approximate nearest neighbor (ANN): HNSW (hierarchical graph), IVF (clustered), DiskANN (disk-based). Plus quantization (PQ, scalar, binary) to shrink memory.
HNSW = current default (fast, good recall). IVF = classic Facebook FAISS-style, good for "cheap search in billions of vectors". DiskANN (Microsoft) = optimised for SSD storage, for datasets that don't fit in RAM. Hybrid search = BM25 + vector + Reciprocal Rank Fusion. Reranker (cross-encoder) as final step.