Whetstone.
All courses
langchainIntermediate

LangGraph Ground School

No toy todo apps. Every module ships a real improvement to a system you use daily: chronicler's classifier, the todoist bridge, Impulse's dice roll, Ark's continuity, cockpit's live panel, vault-tidy's quality bar. You finish able to read any LangChain code fluently, and able to decide, from evidence, where the framework earns its keep in your stack. The capstone rebuilds Impulse itself as a deep agent.

Start the course 18 lessons 900 min
LGCrash course
Curriculum

18 lessons across 9 modules

01Models and Structured Output2 lessons · 45 min
02Tools and the Agent Loop2 lessons · 45 min
03StateGraph and the Dice Roll2 lessons · 45 min
04Persistence, Threads, and Time Travel2 lessons · 45 min
05Human in the Loop2 lessons · 45 min
06Streaming2 lessons · 45 min
07Subgraphs and Multi-Agent Orchestration2 lessons · 45 min
08LangSmith: Tracing, Datasets, and Evals2 lessons · 45 min
09Capstone: Impulse v2, a Deep Agent2 lessons · 55 min
Capstone

Impulse v2, the autonomy engine as a deep agent

The dice roll you love, reborn: checkpointed threads that survive restarts, wild rolls gated by your approval from Discord or the cockpit, live streaming of what Florence’s autonomy is doing, every decision traced in LangSmith, deployed on kinto behind Reception.

Nothing in this module is a new idea. It is modules 3 through 8 wired into one system: the M3 dice StateGraph as the core, an M4 Postgres checkpointer underneath it, M5 interrupt() gating the wild face, M6 streaming the whole run into a live panel, an M7 worker subgraph executing the rolled action, and M8 LangSmith tracing every decision. Production Impulse keeps running untouched; v2 is a parallel build that proves the concepts without risking the real heartbeat.