Labs
Build something real, in under an hour.
Short, hands-on labs with code that runs as-is. Each one ends with a working thing you built yourself — not a slideshow. The fastest way to learn a system is to make it do something.
Labs are quick, single-session builds. For the deep, structured path, see the full courses in Learn.
Available now
6 ready to buildTurn a folder into an OKF bundle, then embed it
Build a tiny Open Knowledge Format bundle (markdown + YAML), then embed it so an AI can find the right file by meaning and answer grounded in its exact contents — the folder-plus-vector architecture, hands-on.
You build:An OKF folder you built, embedded into a vector index, answering questions by finding the right file and grounding its answer in that file — with a citation.
Build a ReAct agent in 40 lines
Write a real Reason+Act agent from scratch — Thought, Action, Observation, loop — that reasons, calls tools, and solves a multi-step question while printing its whole trace.
You build:A running ReAct agent that answers a multi-hop question using tools, with its Thought / Action / Observation trace printed live.
Add memory to an AI agent with LangGraph
Give an agent short- and long-term memory using LangGraph checkpoints, so it remembers across turns and across sessions.
You build:An agent that recalls earlier context within a conversation and persists state between runs.
Put guardrails on an LLM endpoint
Wrap an LLM call with input/output guardrails — rules it cannot cross — so it stays safe and on-topic in production.
You build:An LLM endpoint that refuses or reroutes unsafe and off-policy requests, with an audit trail.
A minimal RAG system in one file
Ground an LLM in your own documents end to end — chunk, embed, retrieve, and answer with citations — in a single readable Python file.
You build:A working retrieval-augmented answer system over your own docs, with sources.
Build your first MCP server in 30 minutes
Stand up a working Model Context Protocol server in Python, expose a real tool, and call it live from an MCP client — no prior MCP experience needed.
You build:A running MCP server that exposes a working tool, connected to a client and answering live.
Want a lab on something specific?
Tell me what you want to build →