Choose how you'd like to sign in
How we started, what we build, and our journey to make AI accessible for everyone.
How we went from handing volunteers a task list to building a system where Ari matches people to meaningful work — and how those volunteers became our most honest source of feedback.
Why we moved from one monolithic server to independent Lambda functions — giving each tool its own isolated brain, eliminating developer conflicts, and opening the door for volunteer contributors.
The Coach — a third AI that reads all your progress and gives honest strategic advice. Born from Reddit posts, a friend suggestion, and the gap between knowing material and passing an exam.
The Extractor — a second AI that silently analyzes every session. How parallel execution crashed our server, back buttons caused race conditions, and why we switched from Sonnet to Haiku for extraction.
We started with simple prompts and got mediocre results. The Socratic method backfired. Students got tired of typing. Mobile keyboards ate the screen. Here is every mistake we made building an AI tutor — and how we fixed them.
Our first data incident — how a storage migration gone wrong taught us about the complexity of adaptive learning infrastructure, and why we now back up daily.
