← Back to the course home

🗓️ The study plan — 13 lessons, 4 weeks

~3 sessions a week: read the lesson (~15 min) + do its lab (~25 min). Tick lessons off — progress is saved in this browser. Watch the companion video before Week 1 for the aerial view.

🐢 Steady: 3 sessions/week → done in 4 weeks.
🐇 Fast track: a weekend per part → done in 2 weekends.
💰 Cost: zero. Plain Python + any free chatbot.
0 / 12

WEEK 1 📖 Foundations — how learning works

The inversion (rules are grown), the red pen, and the puzzle pieces. Labs: two of the three demos.

~40 min
~40 min
~35 min
🏁 Checkpoint: run bigram_model.py and tokenizer_toy.py; explain 'data + answers → rules' and why 'strawberry' is hard — out loud, to someone.

WEEK 2 🗺️ The engine — meaning, guessing, glancing

Embeddings, the next-token game, attention. Lab: the attention demo, twice (once modified).

~40 min
~40 min
~45 min
🏁 Checkpoint: run attention_toy.py; flip 'it' to ball-shaped and predict the change before running; recite query/key/value in school words.

WEEK 3 ⭐ Finishing school & asking well

RLHF/LoRA, prompting on the small desk, and the confident kid. Labs: the A/B prompt test + hallucination hunt.

~45 min
~40 min
~35 min
🏁 Checkpoint: your B-prompt visibly beats your A-prompt; you caught (and verified!) one hallucination; you know when to fine-tune vs RAG — say the rule.

WEEK 4 🧰 Building — open books, hall passes, static

RAG by hand, being the agent's harness, diffusion probes — then the capstone.

~45 min
~45 min
~40 min
~40 min
🏁 Checkpoint: 🏆 Capstone: re-watch the companion video and pause after each of the 9 concepts — explain every one in YOUR school-analogy words. Bonus: build the 30-line RAG from lesson 10 for real.

🎓 13 / 13 — zero magic left, sockets included!

Next: build something — or tour the rest of the school: Kubernetes · Docker · AWS · ArgoCD.