# StudyHub

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Marc Fischer](https://cerebralvalley.ai/u/Aradulei)
- **GitHub:** https://github.com/marcfischer026/StudyHub
- **Demo video:** https://youtu.be/-vJX1kQY6DY
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2713

UniVerse is a study app for university students on Apple Silicon Macs. It turns a semester of lectures into exam practice you can trust.

The problem: two weeks before an exam, the material is scattered. Slides sit in one folder, recordings nobody re-listens to in another, notes are half finished, and the most important sentence of the semester ("this will be on the exam") was only ever said, never written on a slide. General chatbots don't know your course, make up sources, and uploading every lecture, including classmates' side talk, isn't great for privacy.

UniVerse uses two kinds of AI:
- On the Mac: whisper.cpp transcribes lectures offline on the GPU. A local model sorts each transcript into lecture content, questions and side talk, and stars exam hints. Classmates' side talk stays on the laptop. A local tutor answers from the course's own passages with clickable citations.
- Meta's Muse Spark does the one heavy job: it reads the whole course (slides, transcripts, notes, past exams) and writes exam questions, learning objectives and a study guide

The key idea is trust: every claim Muse Spark returns must carry a word-for-word quote from a slide page or transcript line, and UniVerse checks each quote against the source before importing anything. Wrong quotes are rejected and the model is asked to fix them. Students then practise offline with spaced repetition (FSRS), track progress towards the exam, and jump from any answer straight to the slide it came from.

Built in Python with PySide6, LanceDB, PyMuPDF and whisper.cpp, with 809 automated tests.

## More from Meta Global AI Developer Hackathon

- [Kutumb / Family Man](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2710)
- [1Thing](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2711)
- [Pardeep Singh](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2712)
- [4Dimension](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2714)
- [Lift Africa Tech](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2715)
- [VOXEL-SAM AR](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2716)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2713. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt