Architecting and Testing Controllable Agents
summary
Shows how to build agents that trade off flexibility for reliability by constraining control flow with LangGraph, then measures both final answers and tool-call trajectories with LangSmith. Demonstrates a local corrective pipeline—corrective RAG—that grades retrieved docs, falls back to web search, checks for hallucinations, and re-runs generation; reports a 60–80% answer accuracy range and near‑100% tool‑call consistency across models. Includes notebooks for running experiments, tracing timings, and adding in‑app unit‑test style checks.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 1:26Control flow explained
- 8:56LangGraph control flows
- 15:33Agent testing challenges
- 24:40In-app error correction
- 34:04Pre-production testing
- 45:51Production monitoring
- 1:00:22State management
- 1:16:14Local agents and models
- 1:33:02Large tool sets
- 1:41:59Chains vs agents
- 1:51:19Document generation
- 2:00:04Deployment and checkpointing
- 2:11:10RAG challenges
- 2:21:22Q&A
related talks
37mOne agent, one job, better A
25mFrom Chaos to Control: Building Enterprise-Grade LLM Agents That Actually Work
01h 59mBuilding Agentic and Multi-Agent Systems with LangGraph
35mBuilding LLM-based Agents
41mThree agents, three frameworks, one talk
01h 03mAgents Everywhere
31mTesting, Evaluating and Debugging Generative AI and Agentic Ap...
40mBuilding Effective AI Agents Simple, Composable Patterns