Build Story · AI school prototype · Web
AI by AI: learn it by running it
A practice-first AI school. Each lesson opens with a live experiment, and a lab races two models on your prompt.
In one look
The problem
Most people meet AI through articles and demos, then freeze at an empty prompt box. Knowing which model to pick, and when to doubt its answer, comes from trying.
What we built
Each lesson starts with an experiment on two live models, asks what differed, teaches the theory, and ends with a task you turn into your own workflow. An AI navigator answers questions inside the lesson. A separate Model Lab races any two models on your prompt and records which answer you preferred.
Where it is now
A working prototype: 40 lessons in six courses, the Model Lab, sign-in, and Terms and Privacy pages. The free course is free to read. The paid plan is not open; the pricing page takes emails for a waiting list. It has been built and polished, not yet pushed to an audience.
What this would have cost in 2024
About $11k and 6–10 weeks
2024 team: 3 people: full-stack developer, curriculum writer, product designer part-time. Built here by one founder working with AI in 8 working days.
| Piece | 2024 hours |
|---|---|
| Lesson app: courses, modules and lessons, four-step lesson cycle, AI navigator, progress, sign-in | ~93 |
| Model Lab: streaming model proxy, side-by-side race, verdicts, usage logging, daily caps | ~48 |
| Curriculum: six courses and 40 lessons written, structured and linked to further reading | ~83 |
| Site: visual system, server-rendered lesson pages, sitemap, share previews, pricing page, waiting list, legal pages | ~57 |
| Total (PERT) | ~281 |
Method: PERT (O+4M+P)/6 per piece; low = O × $25, likely = PERT × $37.5, high = P × $50 (Accelerance 2024 Eastern-Europe band). These are estimates of what a team would have needed in 2024, not invoices.
How it was built
From first commit to current build in 8 working days
- One evening, a working lesson app
A new Lovable project became a lesson app with sign-in, a four-step lesson cycle, a proxy for live model calls, a waiting list and a pricing page.
13 recorded edits in about four hours - New look, a sixth course, pages search can read
A new visual system and header, a sixth course on running a stack of AI agents, and outside reading on selected lessons. Lesson pages now render on the server with their own titles, a sitemap and real 404s, and Terms and Privacy went live.
45 URLs in the sitemap, 40 of them lessons - Security pass
Hard daily limits on AI use, per signed-in user and across the whole site, and tighter server-side checks on the model calls.
Who did what
The founder decided
- Pro is not open, and the site says so instead of faking a checkout.
- The Model Lab needs a sign-in and a daily cap, because every comparison costs real money.
- Outside material is used when its licence allows reuse, not because it is popular.
AI built and checked
- The web app, built in Lovable from the founder's briefs over 39 recorded edits.
- Course 6, written by Claude from the founder's own playbooks, with Claude checking the live site after each build.
- The October security pass, written with Claude.
Stack
Want something like this for your team?
From one call to something live. The same method is in the Claude Operator Pack.
From the founder
You might also like
Shares TanStack Start · Lovable
Shares TanStack Start · Lovable
Shares TanStack Start · Lovable