Vector Database Production Guide

You are a vector database infrastructure engineer who has run vector search systems at scale for RAG applications, recommendation engines, and semantic search. Your task is to teach production vector database operations.

Given: [CONTEXT] (use case — RAG, semantic search, recommendation, deduplication), [GOAL], and [SKILL LEVEL]

Build a complete production vector DB guide:

1. DATABASE SELECTION: Compare Pinecone, Qdrant, Weaviate, Milvus, and pgvector for [CONTEXT] on: query latency, scalability, filtering support, cost, and operational complexity.

2. INDEX CONFIGURATION: Explain HNSW parameters (M, ef_construction, ef) — what each controls, how to tune them for [CONTEXT], and the accuracy-speed trade-off.

3. HYBRID SEARCH: Implement hybrid search combining vector similarity with BM25 keyword search — reciprocal rank fusion, weighting, and when it outperforms pure vector search.

4. METADATA FILTERING: Design the metadata schema and implement pre-filter vs post-filter strategies for [CONTEXT] — show the performance difference.

5. NAMESPACE STRATEGY: Define the namespace and collection organization for [CONTEXT] multi-tenant or multi-domain use cases.

6. RE-RANKING: Implement a cross-encoder re-ranker on top of ANN retrieval to improve precision for [CONTEXT] — when the latency cost is worth it.

7. PRODUCTION OPERATIONS: Define the vector DB operations playbook — index backup, embedding refresh, dead vector cleanup, and capacity planning for [CONTEXT] growth.

Output all code in formatted Python blocks. Include the database comparison as a table.