Kapa Inspired RAG

Implements a production-grade retrieval-augmented generation pipeline for documentation Q&A. Uses Qdrant for vector storage with Cohere reranking and GPT-4o generation. Includes multi-tenant isolation, freshness handling, JWT authentication, async ingestion jobs, and an MCP server that allows Claude and Cursor to query the knowledge base directly. Validated on 78 Q&A pairs using RAGAS metrics.