Maia
idea· 2026
An assistant to actually learn: it measures what you know in every area, finds the gaps that matter and guides you to close them with methods that work.
Maia maps your level in every area with an assessment that adapts to your answers, finds the most important gaps and guides you to close them with methods research backs: the Feynman technique, retrieval practice, spaced repetition. What you learn ends up in a personal wiki you can browse.
It is the third time I circle the same idea. Cosmoscope started from a question, Upstream from interests; Maia starts from what you know and what you do not, and it is the one that got furthest. The taxonomy covers 19 areas, every screen talks to the real backend and a panel keeps track of model spend.
I first built the whole frontend on fake data, then the backend replaced it one slice at a time, with tests checking the full loop.
Before this
- 2025
Cosmoscope
A question became a concept map built from a web search. One day of work, then it stopped.
- 2026
Upstream
Exploring knowledge on purpose instead of taking the feed: declared interests, chosen sources. It stayed at vision and structure.
