#vector-database

Implement RAG with Pinecone. Architecture: 1. Document chunking and embedding. 2. Store embeddings in Pinecone index. 3. Semantic search with similarity. 4. Metadata filtering for context. 5. Hybrid search (dense + sparse). 6. Retrieve top-k relevant chunks. 7. Augment prompt with context. 8. Generate answer with LLM. Use text-embedding-ada-002 and implement re-ranking for accuracy.

Use ChromaDB for local vector storage. Setup: 1. Initialize persistent client. 2. Create collections with metadata. 3. Add documents with embeddings. 4. Query with similarity search. 5. Filter by metadata. 6. Update and delete operations. 7. Multiple embedding functions. 8. Export/import collections. Runs entirely local, no API needed. Use for privacy-sensitive applications.