You already know a model can talk. Now you make it act. In five 2-hour sessions you’ll build LLM agents that reason in a loop, call real tools, remember, retrieve from your own documents, coordinate as a crew, and ship a guard-railed agent you demo in public.
Start Week 1 →Every week you write real Python in Google Colab, calling live LLM APIs (Claude, GPT, Gemini) through LangChain, CrewAI, and LangGraph. This interactive handbook is the other half of the lab — the mental models, the checkpoints, and the receipts — so nothing about agents feels like magic. Work in pairs (driver + navigator, swap at every checkpoint); your capstone agent is yours alone.
Every agent you meet — a coding copilot, a research crew, a customer-service bot — is some arrangement of these four. An agent is just an LLM wrapped in a loop that can call tools and keep memory. Say the four words out loud.
An agent is a plain LLM given the power to act, observe the result, and try again.
Give it hands and a memory, not just a mouth. Clear tool descriptions are the whole game.
Retrieval beats hallucination. Ground answers in real documents — and cite them.
Split the job, define the roles, let them hand work off. Specialists coordinate; generalists drown.
Log every step, evaluate on real cases, guard the edges. Then — and only then — ship.
Wk 1 — built a ReAct agent that read a question, picked the right tool, and used it in code.
Wk 2 — gave an agent working memory and two tools; it stayed coherent across a multi-turn task.
Wk 3 — built a RAG agent over your own documents that answers with a real citation.
Wk 4 — ran a 2-agent crew with defined roles that handed work off and produced one result.
Wk 5 — added a guardrail, ran an eval, and gave a public 4-beat demo with a shown artifact.
All 5 badges + all 5 missions. Announced at the Showcase.
Where you write and run every Python notebook — no install.
The LLM engines your agents reason with (free/low-cost tiers).
Wire prompts, tools, and models into an agent you can run.
Vector databases that let an agent search your documents by meaning.
Give agents roles and a graph to coordinate multi-step work.
See every step your agent takes — and block prompt-injection attacks.