We were inspired by the hassle of hunting for classrooms on a big campus: especially when schedules live as screenshots in GOLD. We learned to turn unstructured images into structured data, to resolve fuzzy building names, and to give users instant navigation. We built a Next.js app with a minimalist Apple-like UI, a Gemini-powered API that parses schedule screenshots into building/room pairs, and a Mapbox map that drops pins and links to directions. The pipeline: upload is LLM JSON to building resolution to map pin. Challenges: API quotas/model availability, brittle OCR on noisy screenshots, reconciling building names to map data, and keeping UI responsive while handling slow AI calls. In short, we took “Where is my class?” from a guess to a one-tap answer.
Built With
- cursor
- gemmini
- git
- mapbox
- python
- sonnet
- typescript
Log in or sign up for Devpost to join the conversation.