mpc-hacks-2026

Local development

1. Start the python engine socket service:
   • cd python
   • PY_ENGINE_HOST=0.0.0.0 (optional, default 0.0.0.0)
   • PY_ENGINE_PORT=3001 (optional, default 3001)
   • python main.py
2. Start the backend API:
   • cd backend
   • npm install
   • PY_ENGINE_HOST=127.0.0.1 (optional, default 127.0.0.1)
   • PY_ENGINE_PORT=3001 (optional, default 3001)
   • MAP=MAPBOX_API_KEY
   • gemapikey=GEMINI_API_KEY
   • npm run start:dev
3. Start the frontend:
   • cd frontend
   • npm install
   • npm run dev

Health check: GET http://localhost:3000/engine/health

General setup

You can use the docker compose file instead with the command: docker compose up -d --build

For the frontend: GET http://localhost:5173/

For the tuner: GET http://localhost:5555/

Default creds

User: Marc Password: 1234

Note: In PowerShell, set env vars like $env:PY_ENGINE_PORT=3001 before running commands.

Detection strategy:

1. Feature engineering (velocity, cross‑border, night activity, high‑risk category, device/IP sharing, spending patterns).  
2. Rule‑based scoring with weighted heuristics to flag known fraud patterns.  
3. Anomaly detection (Isolation Forest) to catch outliers beyond the rules.  
4. Ensemble verdict that combines rule and anomaly signals into a single fraud score and thresholded verdict.

With another week:

1. Add labeled data and evaluate precision/recall/F1; tune weights and thresholds.
2. Expand signals (geolocation from IP, merchant risk scoring, cardholder behavior baselines).
3. Calibrate explanations (top contributing rules per decision) and improve the UI to drill into why a transaction was flagged.