Supercharging GenAI PostgreSQL Vector Search Optimization
summary
Practical PostgreSQL PGVector tuning for cheaper RAG: recommends starting IVF_FLAT cluster count near the square root of 10k embeddings (≈100) then adjust for accuracy, and tuning list at index build; for graph indexes tune M and construction neighbors and use ef_search at query time. Use LIMIT, FORCE INDEX SCAN, or a distance threshold like distance < 0.3 to cut candidates, and apply categorical prefilters (hybrid search) and choose cosine vs Euclidean per application to reduce memory and latency.
AI-written summary · about
related talks
41mPostgreSQL for AI. Real Vector Search in PostgreSQL
52mpgvector Newbie to Expert
48mFrom Text to Meaning: pgvector Transforms PostgreSQL Search
52mVector search is now boring, but PostgreSQL has ways to go
53mVectors: how to better support a nasty data type in PostgreSQL
47mHacking pgvector for performance
51mVector data in Postgres: Size, TOAST, Filters and Performance
22mComparing vector implementations in generic databases