Project Course
Domain-Specialist Multi-Agent with Per-Agent RAG
Build a supervisor that routes each question to a domain specialist, and give every specialist its own RAG index. The architecture behind the CentenarianOS multi-domain coach.
Course intro video, coming soon
The recorded walkthrough is being uploaded.
Course outline
The supervisor pattern
Route each question to the right domain specialist with a LangGraph supervisor instead of one overloaded prompt.
Per-agent RAG
Give every specialist its own pgvector index, domain-scoped retrieval instead of one shared store that bleeds context.
Retrieval that respects domains
Chunking, embeddings, and metadata filters tuned per specialist so the nutrition agent never answers from the finance corpus.
State and handoffs
Pass context between agents cleanly, shared state, reducers, and handoffs that don't leak one domain into another.
Evaluating a multi-agent system
Use LangSmith traces and datasets to score the system per-agent, so a regression in one specialist is visible before it ships.
Shipping it
LangSmith Deployment, Drizzle-backed persistence, and the path from notebook to the live Fit T. Cent coach.
Tools used
What this course rests on
The literature behind the multi-agent RAG patterns taught here.
- Lewis, P., et al. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. https://arxiv.org/abs/2005.11401
- Karpukhin, V., et al. (2020). Dense passage retrieval for open-domain question answering. https://arxiv.org/abs/2004.04906
- Yao, S., et al. (2022). ReAct: Synergizing reasoning and acting in language models. https://arxiv.org/abs/2210.03629
- Asai, A., et al. (2023). Self-RAG: Learning to retrieve, generate, and critique through self-reflection. https://arxiv.org/abs/2310.11511