Build Log / Mar 9–10
Week 0: The Build
Shipped the entire 9-lesson course in two days. Pyodide, React Flow, trace pipeline, deployed on Railway.
Two days, nine lessons
The idea was simple: teach AI agents by building one from scratch. No framework, no install, everything runs in the browser. I wrote all 9 lessons in two days — from "an agent is a function" to "the whole thing in 60 lines." Pyodide handles Python in WebAssembly. React Flow renders animated architecture diagrams. A custom trace pipeline shows every HTTP call and tool dispatch in real time.
V2 redesign on day 2
The first version used Mermaid.js for diagrams. They looked fine but couldn't animate. Ripped it all out and replaced with React Flow — now the diagram highlights nodes as code executes. Same day: added the trace parser, the monitor panel, the step-by-step prose column. Deployed to Railway by end of day.
What I shipped
9 lessons covering: agent function, tool calling, the agent loop, conversation history, state tracking, persistent memory, guardrails, self-scheduling, and the full agent composed in ~60 lines. Each lesson has interactive code you run in the browser, a live architecture diagram, and a monitor showing the raw HTTP traffic. Mock LLM mode for people without API keys.