Spotted Monorepo

Stack (MVP)

• Backend: FastAPI (Python), Postgres + pgvector for vector search
• Embeddings: OpenAI text-embedding-3-small (swappable later)
• Infra: Docker Compose (Postgres + pgvector)
• Mobile: React Native (Expo) under apps/mobile

Structure

• apps/api: FastAPI service and recommendation logic
• apps/mobile: React Native app (Expo)
• infra: Docker compose and ops
• .env.example: environment template

Quickstart

1. Copy envs

   • Create .env from .env.example and fill values (OpenAI key, DB URL).

2. Start database (Postgres + pgvector)

   • docker compose -f infra/docker-compose.yml up -d

3. Backend (FastAPI)

   • cd apps/api
   • python3 -m venv .venv && source .venv/bin/activate
   • pip install -r requirements.txt
   • uvicorn app.main:app --reload

4. Mobile (React Native, Expo)

   • Bootstrap later: npx create-expo-app apps/mobile
   • Then run: cd apps/mobile && npm start (or npx expo start)
   • Recommended libs: react-native-gesture-handler, react-native-reanimated, expo-location

Recommendation Stack (MVP)

• Compute embeddings with OpenAI text-embedding-3-small (1536 dims) for users and items.
• Store vectors in Postgres using pgvector (vector(1536) column).
• Retrieve candidates with ANN index (ivfflat or hnsw) and re-rank in Python by:
  • cosine similarity × distance_decay × time_decay (for events) × community_boost × popularity_prior.
• Log swipes (save/pass for events, connect/pass for people) to improve weights.

Notes

• UCLA-first: seed event ingestion with campus calendars + 1–2 city sources.
• Keep embeddings behind a thin interface to swap to local models later.