ProtoRyde Backend

Feature-complete backend repository for the hackathon.

Structure

• app/ -> active FastAPI backend
  • api/ endpoints
  • core/ database models + scheduler
  • triggers/ weather + fraud trigger logic
  • services/ pricing, ML+SHAP prediction, PDF generation, model training utility
  • tests/ contract tests
• docs/API_CONTRACT.md -> frontend/backend integration contract
• docs/BACKEND_RUNBOOK.md -> backend owner runbook + curl payloads + error handling
• main.py -> root ASGI entrypoint (app.main:app)
• requirements.txt -> single dependency source

Run

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

ML + PDF now integrated in active app

• ML status: GET /api/premium/model-status
• Premium prediction (ML-first with fallback): POST /api/premium/predict
• Policy PDF download: GET /api/policies/{rider_id}/current/document
• Optional local model training: python app/services/train_model.py

Demo-Critical Endpoints

• POST /api/demo/bootstrap
• POST /api/triggers/simulate
• GET /api/claims/{rider_id}
• GET /api/policies/{rider_id}/current
• GET /api/policies/{rider_id}/current/document

Reliability Tests

python3 -m unittest app.tests.test_phase2_contracts -v
python3 -m unittest app.tests.test_api_endpoints -v

Deploy on Render

This repo includes render.yaml and Procfile for Render deployment.

1. Create a Blueprint service in Render and point it to this repository.
2. Set service root to protoryde-backend if deploying from the umbrella repo.
3. Render provisions protoryde-db from render.yaml and injects DATABASE_URL.
4. Keep ENABLE_SCHEDULER=false in production unless you intentionally want live polling.

Beginner flow (recommended)

1. Run preflight locally:

   ./scripts/render_preflight.sh

2. Deploy from Render Blueprint.
3. Run live smoke checks against Render URL:

   BACKEND_URL=https://<your-service>.onrender.com ./scripts/render_live_smoke.sh

See full guide: docs/RENDER_FIRST_DEPLOY.md

Notes

• The app normalizes postgres:// to postgresql:// automatically for SQLAlchemy compatibility.
• model.pkl is optional; premium prediction falls back to the rule engine when the model is absent.