LLM Explainer
An interactive LLM learning tool built from Stanford CME 295 lectures — building it forced me to understand each concept, not just consume it.
The problem
I enrolled in Stanford's CME 295 (LLMs for PMs and Executives) to learn how LLMs actually work — not at a surface level, but well enough to make good product decisions about them. Reading the slides and watching lectures wasn't clicking fast enough. I needed something I could interrogate, not just consume.
My hypothesis
If I built the learning tool myself — forcing me to explain each concept clearly enough to code it — I'd understand the Stanford material faster than reading it.
What I built
All content is sourced from Stanford CME 295 — lecture notes, slides, and transcripts — with full attribution in the header. Six concepts from the course — tokenization, embeddings, transformers, training, RLHF, agents — each with an interactive playground you can actually poke at. Saves to your iPhone home screen as a standalone app. A floating recorder lets you log questions by voice as you go, tagged to whichever concept you're on. One tap sends everything to ChatGPT, Notes, or Messages; download as a text file if you're on desktop.
What broke
The share button is built and shipped but not yet verified on a real iPhone — there's no way to test iOS-native share sheets from a laptop. Voice recording may also be unreliable when the app is running in home-screen mode, which is a known iPhone limitation. The one-tap export is the workaround.
What I learned
Building the explainer forced me to understand each concept well enough to write interactive examples — that's a different level than reading a paper. The question recorder revealed something: the moments I got stuck were the most valuable ones. Logging them turned the stuck moments into a record of where AI education actually breaks down.
If I kept going
Verify the share sheet on a real iPhone. Add more concepts as the CME 295 course progresses. The question archive is already a record of where AI education actually breaks down — a PM's field notes on learning each concept from scratch.
Prototype
The Text-as-Numbers playground: sentence → tokens → IDs → embeddings in meaning space
Mobile hero — saves to your phone home screen as a standalone app
Full walkthrough: six concepts from Stanford CME 295, each with an interactive playground