catch fever

a dating app for unc students that matches people on personality, not looks. instead of endless swiping, everyone gets a small set of matches once a week.

🫟 in early development.

how it works

1. verified signup. only @unc.edu emails can create an account (supabase auth).
2. personality profile. users answer prompts about values, interests and how they spend their time.
3. ai matchmaking. answers are embedded with sentence-transformers, enriched with claude haiku 4.5, and pairs are ranked with a lightgbm model.
4. weekly match drop. a supabase pg_cron job releases new matches each week, and an expo push notification lets users know.

tech stack

layer	tools
mobile app	typescript, react native, expo, expo router
backend	supabase (postgres, auth, edge functions, pg_cron)
matching	python, sentence-transformers, claude haiku 4.5, lightgbm
notifications	expo push notifications

architecture

signup (@unc.edu) → profile answers → embeddings + trait extraction
      → lightgbm pair ranking → weekly pg_cron drop → push notification

project structure

catch-fever/
├── app/        expo / react native app
├── supabase/   migrations, edge functions, cron jobs
├── ml/         matching pipeline
└── docs/       architecture notes and decision records

getting started

requirements: node.js lts, python 3.12, and expo go on your phone (or an android emulator).

cd app
npm install
npx expo start

scan the qr code with expo go, or press a to open the android emulator.

copy .env.example to app/.env and fill in your own keys. never commit .env.

roadmap

• project scaffold
• connect app to supabase
• @unc.edu signup and email verification
• profile and personality prompts
• embedding and trait extraction pipeline
• lightgbm ranking model (trained on synthetic data)
• weekly match drop with pg_cron
• push notifications
• chat between matches

disclaimer

this is an independent student project. it is not affiliated with, sponsored by, or endorsed by the university of north carolina at chapel hill.