How this course was made

A course about models that make judgments, built with models that make judgments. Here's the pipeline, so you can decide what to trust.

1. Sources. We collected the TypeSafe, OpenAI, Cloudflare and AWS documentation and cookbooks, 58 talks and podcasts (transcripts fetched in English), 18 papers, and articles, benchmarks and Hacker News threads from independent testers: 199 sources in all.
2. Distillation. A model read every source and extracted specific, citable insights, each with a verbatim quote (and a timestamp for talks). Quotes were automatically checked against the source text.
3. Measurement. We ran our own experiments (14,318 API calls) so the course wouldn't depend on vendor numbers: see what we measured. Every answer shown on screen in a lesson is a real answer, labeled with the model that gave it.
4. Writing. Lessons were written by Claude from those insights and measurements, and every number and quote was checked against its source before narration.
5. Animation. Lessons are not video files. Each one is a list of scenes rendered live by a small JavaScript engine, synced word by word to the narration, with motion rules borrowed from character animation: springs by feel, a camera that only moves for a reason, and a stepped cast.
6. Narration. The voice is an AI voice (ElevenLabs).

Mistakes are possible. If you spot one, open an issue with the lesson and timestamp. The engine, lessons, experiments and build are all on GitHub.