AAPD, an education platform for product builders in Taiwan, runs a Career Coach Program: one-on-one mentorship with 24 coaches. People struggled to pick one, and third-party quiz tools came with ads and limits. So I built a matching quiz, with AI writing all the code. It went live in 18 hours, working one hour a day.
I gave Claude Code the old tool and the context. The first version worked, but it did not look like AAPD. Fixing that through prompts alone took too much back and forth, so I changed the workflow:
The team needed to keep it running without me. Firebase was an option, but no one on the team knew how to maintain it. So I kept it simple: the coach list lives in a Google Sheet, and a short Apps Script sends it to the quiz. To add or update a coach, you just edit the sheet. The site itself is hosted on GitHub Pages, which is free and easy to manage.
Five AAPD team members tested it before launch. The flow felt smooth, but two gaps came up: it was Chinese only, and nobody had checked it on mobile. I fixed both before shipping and added tracking to measure bookings.
It now works on any screen size, has light and dark mode, and supports Chinese and English. It also comes with a token library and a markdown guide for the next project.
Any screen size. The layout works from desktop all the way down to mobile.
Light and dark. Switch the theme in one tap.
Chinese and English. Testing showed it was Chinese only, so the whole quiz now switches language in one tap.
The real decisions were not about prompting. They were about knowing when to step in: choosing Figma, judging the matching logic, picking Google Sheets over Firebase. AI handled the execution. I handled the judgment. It also pushed me to learn some basic coding, so I can fix the obvious things faster myself.
