Multi-Agent LangGraph with Per-Agent RAG

by bam

Build one real multi-agent coach end to end, a LangGraph supervisor that routes each health question to the specialists that can actually answer it. Every specialist owns its own retrieval namespace, so a nutrition query never wades through workout documents, and each answer carries citations back to the snippets it was built from. You'll design the routing schema with Zod, guarantee ordering through the graph's shape, and isolate specialists by type so none can read another's findings, then measure routing, citation coverage, and grounding with LangSmith evals that gate every change. It's grounded in the open-source Centenarian Coach repo, TypeScript first with a Python translation track, and written to code along.