An expanding interactive course for software developers who use AI tools and want to understand the machinery underneath them. The coding agent is the primary adaptive teacher; the adjacent browser is the workbench for predictions, explorables, evidence, progress, and optional reference notes. The complete nineteen-lesson foundation begins with a six-lesson AI 101 terminology runway, separates training, inference, generation, and reasoning, then builds from scalar gradients through a trained tiny Transformer, cached autoregressive generation, sampling, and claim-aligned evaluation.
Basic programming, ordinary arithmetic, arrays, a terminal, Git, and the ability to read a test failure. No machine-learning or calculus background is assumed; TypeScript is introduced where the exercises need it. The current course takes roughly twenty hours.
From the repository root:
pnpm install --frozen-lockfile
pnpm courseOpen the printed URL in the Codex built-in browser or Claude Code Desktop Preview pane. Course prose is CC-BY-4.0; code is MIT licensed. Contributions follow the root CONTRIBUTING.md.