Just a few more things that didn't make it into the post:

Stack: Next.js frontend, hosted on Vercel. Course curriculum generation and diagnostics done via LLM calls - I'm not planning to open source the prompt/architecture right now, but happy to dive deeper on specific aspects if people are interested.

Why I built it: I realized that the real failure point for most forms of self-taught learning isn't lack of content - it's that consumption of the content feels like learning even when you're not retaining anything. You watched this lecture or this video? You got it. I wanted something that wasn't complete until you'd proven it.

Limitations of what I've built so far, just to be clear:

AI Viva (voice defense) isn't out yet - it's the next big feature I plan to work on The curriculum graph generator works well enough for structured domains (CS, math, econ), but isn't great for qualitative subjects yet No mobile app, only web version right now

Pricing: Free at the moment, no paid tier yet. No idea how I'm going to monetize it yet either.

Privacy and Security: Supabase with row level security such that users will only ever be able to access their data via the DB layer, and all access is via tRPC procedures (nothing client side), and there's rate limiting in place on the API as well. Once the voice version of Viva comes out, I'll write up details on how voice data is stored.

I would truly appreciate any and all criticism, particularly from anyone who has attempted something like this "prove it" system in the past (the Recurse center, viva based grading systems, etc.).