Day 4: Multi-Agent Systems and Human-in-the-Loop

When one agent isn't enough: agentic retrieval, orchestration patterns, shared state between agents, approval gates and escalation, and how multi-agent systems fail.

Topic Description & Objectives Resource
Multi-Agent Systems and Human-in-the-Loop From RAG to agentic RAG: query analysis, self-correction and multi-step retrieval; router, supervisor-worker, peer and hybrid orchestration; shared state, structured messages and cost; the trust spectrum, approval gates and escalation; cascading errors, loop detection and mitigation Coming soon
Lab 1: Agentic RAG and Multi-Agent Patterns The Day 4 lecture as runnable code: a query analyzer and self-correcting retriever, router, supervisor-worker and writer-editor patterns, shared state and the context-loss bug, a risk-based approval gate and escalation, and a loop detector, wired together in a small support system. Coming soon
Lab 2: A Team of Agents Triage, two specialists, a writer, a reviewer who is not the writer and a supervisor that enforces a budget and hands cases to a person; then compare the team against one agent with all the tools on cost and correctness. Coming soon