No videos. No lectures that outlast a demo. In five 2-hour sessions you’ll train neural networks in real Python, do the learning math by hand, build a language model from scratch, measure real bias, and ship an AI project you demo in public.
Start Week 1 →You’ll write real Python in Google Colab every single week. This interactive handbook is the other half of the lab — the concepts, the checkpoints, and the evidence — explained step by step so nothing in AI feels like magic. Work in pairs (driver + navigator, swap at every checkpoint); your capstone is yours alone.
Every single thing you build in the next five weeks is some version of these four. Say them out loud.
A model is only as good, or as fair, as what you feed it.
Bias comes from data, not villains. Measure per-group or you’ll never see it.
A model is never lying and never honest — it can’t tell the difference. Checking is your job.
On demo night and on your resume, forever. When in doubt, cut.
Wk 1 — trained a classifier in code, reported accuracy from a proper split, tuned two knobs.
Wk 2 — did 3 gradient-descent steps by hand, explained the U-curve, measured recall-for-8.
Wk 3 — built a working text generator with a temperature dial, logged a red-team technique.
Wk 4 — scoping canvas complete, GREEN LIGHT earned, build started.
Wk 5 — public 4-beat demo with one shown artifact; a resume bullet with a number.
All 5 badges + all 5 missions. Announced at the Showcase.
Where you write and run all your Python notebooks — no install.
Watch a neural network learn, live, in your browser.
A legal prompt-injection game for learning to red-team AI.
LLMs (free tiers) — your explain-only lab assistant.
Turn your model into a shareable web demo.
Pose landmarks for the “Coach’s Eye” capstone.