Product engineer. I ship the whole thing, and I build it for the person on the other side of the screen.

Long technical docs lose me halfway through. I wanted explainers built for a short attention span, where every choice about structure comes from research on how people actually remember things, not from taste. And I wanted an AI to write them in that format for me, about anything, including our own codebase.

• A Claude skill that researches a topic before writing, reading the real files when the topic is a codebase, then writes the doc as plain markdown.
• A reading app that renders every doc the same way, so changing how all of them look is one stylesheet, not forty files.
• A graded quiz of real-world scenario questions, one at a time, with an explanation for every answer and a score at the end. It is built on active recall.
• Dark mode, three reading widths, and a 'bionic reading' toggle that bolds the start of every word, all remembered per browser.
• On a phone, the table of contents folds into one sticky bar that shows which section you are in.

• A beautiful page around wrong facts is worse than no page, so the skill researches before it writes a single heading.
• Keeping the content as plain markdown and every visual decision in the app is what lets forty docs change their look at once.
• An automated check on every doc's quiz catches what a human reviewer skims past.

• Designed the doc format first: a short header, small sections, callout boxes for the key idea and the common trap, and one quiz block at the end.
• Built the reading app with React, Vite and Tailwind, with syntax-highlighted code and collapsible deep dives.
• Packaged the writing process as a skill anyone can copy into their own AI setup.
• Wrote a content checker that runs on every change and blocks a doc with a broken quiz or a missing section.

Every push runs the checks, builds the site, and publishes it to GitHub Pages with GitHub Actions.