vector database indexing

CLI-first guide to optimizing RAG for accuracy, latency, and cost: choose and tune the right index (FAISS/Chroma/Qdrant/Milvus; IVF/HNSW), chunk smarter (256–512 tokens, light overlap), speed embeddings with batching/threads/GPU/caching, improve recall with hybrid BM25+dense and light reranking, and measure recall@k, p95 latency, and cost via FastAPI+ab and ragas—complete with multi-distro installs and reproducible scripts.