Why this matters for PMs right now
Two years ago, "AI PM" was a specialisation. Today it's closer to a baseline expectation — the way "can query data in SQL" became table stakes in the 2010s. If your product touches a recommendation, a chatbot, a search box, or anything generating text, you're already making AI product decisions. The only question is whether you make them with a real mental model or by vibes.
This isn't about becoming an ML engineer. It closes a specific, common gap: PMs who talk fluently about "AI strategy" in the abstract but go quiet the moment a question gets concrete — why is this feature slow, why did the model get worse after the last update, why does it confidently make things up, why is this multilingual feature costing 3× more than expected. Those are product judgment calls dressed up in technical vocabulary.
Examples are grocery-delivery flavoured so the concepts attach to something concrete — swap in your own domain when you present.
Course map
Four chapters, building one connected mental model. Start at the top, or jump to a topic.
ML Foundations
Neural nets, overfitting, gradient descent, loss curves, data splits, distribution shift, early stopping, hyperparameters, the pre-ship review.
How LLMs Work
Tokenisation, counting letters, context windows, attention, RAG, latency, embeddings, temperature, hallucination, inference, tool calling.
Evals, Safety & Cost
Evals, benchmarks & Goodhart's Law, the eval harness, AI safety & alignment, and the environmental impact of LLMs at scale.
Watch & Apply
Two recommended deep-dive talks, plus a cross-chapter capstone that ties the whole course together with a 10-question self-check.
Three revise modes
Quick keyword lookup (deeplinkable), flashcards by chapter, or a shuffle-able run through all 25 for interview prep.
Final assessment
A mixed cross-chapter quiz with an optional shareable result. The "can I use this for real?" check.
Author & contributing
What this is, who it's for, the licence, and how to suggest improvements or new concepts.
Suggest & vote on topics
Tell us what to cover next and upvote the ideas you want most. The top-voted topics get built — and marked shipped so you can watch it happen.
How to use it
Read each concept top to bottom: the core idea in one sentence, a short plain-English explanation, an interactive visual to play with, then the PM takeaway and the interview line worth memorising. Drag the sliders and click the tabs — the visuals are where the intuition lands. Pass each chapter quiz (80%) to mark progress; it's stored only in your browser.
No login required, nothing to install — the course itself runs entirely in your browser. The only data stored anywhere is what you choose to post on the community board (anonymous). Safe to share with anyone — fork it and adapt it for your own team.