RAG-powered customer support assistant using our docs

We want to build a RAG-based support assistant trained on our 500-page documentation. The assistant should answer accurately with citations, escalate to human agents, and integrate with our existing chat widget.

You should be comfortable with LangChain or similar, vector databases, and LLM API integration. A working demo first, then production hardening.