Contextual search with vector search: exploring your options with open source tools
summary
Demos convert 35,000 movie plots into 512‑dim vectors with the TensorFlow Universal Sentence Encoder and store them in Postgres + PGVector for brute‑force KNN and indexed ANN; images use OpenSearch + OpenAI CLIP embeddings; recipes use a 384‑dim Hugging Face model loaded into ClickHouse for fast analytical search. Covers index tradeoffs (IVF_FLAT vs HNSW), distance metrics (L2, cosine, inner product), filtering strategies, and a RAG pipeline using LangChain + AWS Bedrock for retrieval plus generation.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:34What is vector search
- 5:28Movie recommendation demo
- 17:39KNN vs ANN indexing
- 25:01HNSW vs IVF comparison
- 32:04Multimodal search demo
- 36:08Distance metrics
- 40:50Vector search with filtering
- 42:14ClickHouse for recipes
- 47:10RAG with vector search
- 54:58Embeddings tips
- 57:02Resources and next steps
related talks
30mVector Search Made Simple: Getting Started with OpenSearch for AI Applications
48mFrom Text to Meaning: pgvector Transforms PostgreSQL Search
27mMachine Learning for Advanced AI Search
50mA fun and absurd introduction to Vector Databases
19mVectorize Your Open Source Search Engine
46mNatural Language & Search, Improve your Search w/ Vector capabilities
36mUnleashing the Power of Vector Databases for AI-Driven Applications
25mBeyond Filters: Modern Search with Vectors in Django with Kumar Shivendu