John McGlone

AutodemiaBuilt end to end2026

An AI study partner, with 260+ merged pull requests in week one.

Autodemia is an AI study partner that remembers what you’ve learned. Every subject becomes a concept map, and what you master in one counts in the others. I designed it, built it, and shipped it myself: I wrote the specs, Claude Code wrote the code, and I reviewed every change.

The product · Autodemia

Problem
Study tools treat every subject as a fresh start and quiz you on a fixed schedule. They don’t know what you already know.
What I built
The whole product: the Nuxt web app, Postgres, Stripe billing, web push, and the AI that teaches, quizzes, and cites your sources. Mastery is tracked with an Elo-style score and spaced repetition.
Stack
Nuxt 4Postgres + pgvectorDrizzleClaude APIStripeWeb push
Speed
First commit October 28, 2026. 260+ merged pull requests in the first week.

Result

  • Live at autodemia.io
  • 260+ merged pull requests in the first week
  • More than 1,000 automated tests gate every merge
  • Model routing and credit caps keep AI spend below revenue

Problem

A whole product with no team to split it across: web app, database, billing, and the AI, all on one person.

My part

Everything: product, design, code, billing, and the AI. Claude Code wrote the code from my specs, and I reviewed every pull request.

Hard call

Start every AI call on the cheapest Claude model that fits, and step up only when you correct it. Credit caps keep AI spend below revenue.

Result

Live at autodemia.io, with 260+ merged pull requests in its first week and more than 1,000 tests gating each one.

  1. A map, not a deck

    Each subject is a concept map. Mastery is tracked per concept with an Elo-style ability score, and spaced repetition brings a review back right before you’d forget it. What you master in one subject counts in the others.

  2. Practice for what you perform

    Recall isn’t the whole job. Autodemia has a speaking coach, mock interviews, and a whiteboard with AI critique, and its answers cite your own sources.

  3. How it shipped that fast

    I write a spec. Claude Code builds it in a pull request. Lint, type checks, and tests have to pass before I review it and merge. That loop ran more than 260 times in the first week.