Step by step to your own agent.
From the first install to a productive setup. Every tutorial will get a video and cheat sheet over time.
Get an OpenRouter key and make your first call
Create an account, generate an API key, fire off your first call with curl and an SDK - and understand what's in the response.
How neural networks learn to read - the MNIST lab
Draw digits, train a real neural network live in your browser, and watch it learn to recognize handwriting from 784 pixels - from scratch, no server.
A Soul in a File - what belongs in a soul.md (and what doesn't)
Same world, different soul. Ask the same question to agents that differ only in their soul.md - none, minimal, rich, overloaded - and see the effect instantly.
Hugging Face for beginners
Hub, Spaces, inference - eight chapters from "I know the download button" to a publicly reachable demo with a GPU. No server, no credit card, no graphics card of your own.
Understanding RAG - step by step with real data
We build a Retrieval-Augmented Generation system from scratch. Real embeddings, a real vector database, and demos you can pause and pull apart whenever you want.
Build your own PyTorch
Tensor, autograd, training loop. Built by hand in eight steps, with NumPy and without torch. By the end your own framework separates a dataset with over 90 percent accuracy.
Build your own tiny agents system
Concierge and runner built by hand - a router without a single LLM call, an agent with a fresh context, a hard compression boundary. Eight steps against any OpenAI-compatible endpoint, no framework and no vector database.
Build an MCP server that does something
No echo tool: a server with a domain, nine tools, two transports and two auth models - connected to a real client and deliberately made to fail.
Fine-tune FunctionGemma on your own MCP
A 550 MB model learns the eleven tools of the sim worlds. Three measurements on one test bench, a distilled dataset and a LoRA run that fits on a small card.