Inspiration

This project is inspired by spaced repetition apps supermemo and anki that help people study and retain knowledge

What it does

It solves a pain point most students have about flashcards: they're time consuming and annoying to make. This app extracts key insights from your study material and automatically creates flashcards. The cards you get right appear less frequently so you naturally spend more time focusing on what you need to study.

How we built it

We used an obsidian plugin as a base because it had much of the needed typescript components already implemented. We built it into a web app, hosted it on vercel and set up nemotron on an Nvidia-hosted Blackwell GPU to handle automatic extracts and flashcard generation. We used tiger database for the backend.

Challenges we ran into

We got mixed up with the Nvidia endpoints (integrate vs brev) and weren't sure which launchable to use. When using ai agents sometimes they did things that set us back. We took production down once by shipping code ahead of a migration that had only ever been applied by hand, which is why the build now runs migrations before it builds anything. Vercel cost us an afternoon on its own: vercel link overwrote our .env.local with no backup, and vercel ls --prod does not list newest first, so we redeployed an old commit and spent an hour chasing a bug we had already fixed

Accomplishments that we're proud of

We're happy to have completed an MVP that integrates AI and can be showcased on any device.

What we learned

Check what your deploy tool actually does before trusting it. We assumed vercel ls --prod listed newest first and lost an hour to that.

Sometimes documentation just isn't good enough and instead of trying to Ctrl+f for answers, you just gotta experiment and find what works.

We also spent a lot of time dividing up the work and settling on feature branches so our efforts wouldn't collide with each other, which ended up saving us lots of time.