Loading...
Loading...
Stop stuffing every tool into one ReAct loop. Build a LangGraph supervisor that classifies intent, dispatches to a specialist subagent with its own prompt and tools, streams each event with clear agent attribution, and remembers the conversation across turns.
Message a mentor about fit, prerequisites, or where to start. Replies come on WhatsApp, usually within a day.
Engineers are learning here from
Build a supervisor-routed multi-agent assistant in LangGraph. A cheap router classifies every user message into one of four specialist subagents (RAG, SQL, hybrid, human handoff) that share typed state, stream per-agent events over SSE, and remember prior turns through a checkpointer.
Route user messages to specialist subagents with a LangGraph supervisor and stream each one over SSE.
What you'll ship
What you'll learn
Curriculum
Knowledge base ingest
Chunk policy documents into ChromaDB with layout-aware splitting so the retrieval subagent can cite real sources
Structured data layer
Design a tiny SQLite schema for customer and claim records so transactional subagents answer from real rows
Shared typed state
Define the TypedDict every node reads from and writes to, so routing, retrieval, and answers stay clean across subagents
Specialist subagents
Build the four specialists: RAG-grounded, SQL-grounded, hybrid, and human escalation
Supervisor router
Build the supervisor node that classifies every user message and dispatches to the right specialist
Streaming endpoint with SSE
Expose the graph over FastAPI SSE and attribute every chunk to the subagent that produced it
Multi-turn memory
Persist conversation state with a LangGraph checkpointer keyed on thread id so follow-ups stay coherent
Who it's for
who have shipped a single-agent chatbot and now see it fail on messages that span policy, data, and human handoff
who want the supervisor plus specialist pattern without reading the entire LangGraph source tree first
adding an AI copilot to an existing product and need clean separation between a knowledge base, a transactional database, and a human fallback
FAQ
Yes, comfort with StateGraph, nodes, and edges is expected. The prerequisite workshop on LangGraph multi-agent workflows covers the fundamentals. This course goes deeper into the supervisor pattern, shared state design, per-agent streaming, and checkpointed memory.
The workshop uses a provider pattern that works with Fireworks, OpenRouter, Gemini, or OpenAI. You pick one provider in .env and every subagent, including the supervisor, uses it. Fireworks has a generous free tier that is enough to finish the course.
No. The reference app happens to be an insurance copilot because it has a clean split between a policy knowledge base, a customer database, and an escalation path. The pattern applies to any domain where one assistant needs to answer from documents, from a database, and know when to hand off to a human.
Yes. The RAG helpers live behind a small interface. Once the graph is wired, pointing the retriever at Qdrant, Pinecone, or PGVector is a single file change.
Pricing
Subscribe to Pro for every paid course, or buy just this one.
Unlock this course and every paid course plus workshop replays. One subscription.
You save 54% with regional pricing
One-time purchase. Lifetime access to every lesson, exercise, and update.
You save 47% with regional pricing
Still deciding? Ask Param a question
Supervisor-routed multi-agent systems with LangGraph
$79 one-time