Build AI agents thatactually do things.
MCP, multi-agent systems with Pydantic AI, and local models with Ollama. Coded live in one Colab notebook.
$ manager.run("Plan a weekend trip to Kannur and book a resort.")
[MANAGER → RESEARCHER] weather in Kannur + real hotel names?
[Research] check_weather("Kannur") → 84°F, clear
[Research] search_web("resorts in Kannur") → 3 results
[MANAGER → BOOKER] book "Blue Bay Resort", this weekend
[Booking] book_hotel(confirm=True) → CONFIRMED
FINAL: Trip planned. Blue Bay Resort booked for Sat–Sun.
Watch a free lesson
Lesson 8, unlocked. The math that shrinks a 4GB model to 1GB.
Why this course
Whiteboard First
Every concept drawn by hand before any code appears.
No API Bills
Runs free on Colab with Ollama and Gemma 4 E4B. No keys, no per-token costs.
One Notebook
Every coding session in a single Colab. Run it top to bottom.
Production, Not Toys
Ends with a real REST API: FastAPI, JWT auth, hashed passwords.
Ten lessons. One working system.
Each lesson builds on the last.
Phase 1 · Protocol
The wire format under every agent framework.
Phase 2 · Agents
A local brain plus MCP hands.
Phase 3 · Model Internals
Tokens, bits, and quantization.
Phase 4 · Production
The capstone, coded live.
What you leave with
Speak MCP Fluently
Write MCP servers and tool schemas by hand.
Build Multi-Agent Systems
Orchestrate multi-agent, multi-tool systems.
Run Models on Your Metal
Pick the right quant and run models locally with Ollama.
Ship a Secured Agent API
Ship a JWT-secured agent API.
Frequently asked questions
Stop prompting. Start building agents.
One payment. Lifetime access. Every future update.
One-time payment · no subscription · all future course updates included