A full trading-education app, idea to working product in one morning

The biggest number comes first, with its method underneath. Every figure is read from the build's own git log or a dated usage reading, and nothing here touched anyone's private data.

How fast

The idea came up at 05:10 and the first commit landed at 05:41. By 08:40 the app had 56 commits and 14 features, each built on its own branch and merged into one product. That is about one commit every three minutes.

What it cost

No extra spend. The build ran inside a flat-rate AI subscription that was already paid for (about US$200 a month), and it used about 6% of one week's allowance, going by the usage meter readings before and after. Pro rata, that is roughly US$3 of the subscription. Hosting runs on Cloudflare's free tier, so the running cost is $0.

What was built

• A 200-lesson trading course with server-scored quizzes, progress and badges. Each lesson was checked by a second, separate fact-checking agent.
• An AI tutor on every screen, from a floating chat bubble that knows what you are looking at (a lesson, a chart or a practice round).
• Live coin and currency charts with drawings, indicators and a practice-money trade journal.
• A Telegram bot: account linking, lesson reminders and signal alerts with one-tap practice trades.
• A gamified garden that grows from learning habits, with levels, seasons and streaks.

How it was checked

Every change had to pass the test gate before it landed. Then a headless phone browser tapped through every tab at two screen sizes, opened the AI tutor and checked for a real answer. That pass caught two layout bugs (overlapping chat buttons and an overlapping price label), and both were fixed in the code the same morning.