Sentinel is a multi‑agent fraud intelligence platform that scores real‑time events, generates alerts and cases, and provides an analyst console for investigation. The AI model stack is branded Aegis.

• Ingests fraud‑relevant events (transactions, logins, device changes)
• Scores risk using multiple signals and fuses them into a final decision
• Produces alerts and cases for analyst workflows
• Visualizes activity in a web console (Command Center, Live Transactions, Cases, Graph, Timeline)

# From repo root
docker-compose up --build

Services:

• Frontend: http://localhost:3000
• Backend: http://localhost:8000

# Backend
cd backend
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

# Frontend (new terminal)
cd frontend
npm install
npm run dev

• POST /api/events — ingest an event
• POST /api/score — score an event through the Aegis pipeline
• GET /api/dashboard/feed — latest alerts for the UI
• GET /api/dashboard/metrics — KPIs for the Command Center
• GET /api/cases / POST /api/cases — list/create cases
• POST /api/demo/seed — generate demo alerts
• POST /api/demo/reset — wipe demo data for the tenant

1. Open Command Center and click Seed demo data.
2. View alerts and risk scores in Live Transactions.
3. Explore linked entities in Fraud Network.
4. Open a case from the Cases page.

• backend/ — FastAPI app, scoring pipeline, services
• frontend/ — React console UI
• services/ — auth, scoring, ingestion, case management
• database/ — SQLAlchemy models and DB wiring
• docker-compose.yml — local stack (Postgres/Redis + app)

Environment variables (Docker uses docker-compose.yml):

• DATABASE_URL — PostgreSQL connection string
• MAFI_OPENAI_API_KEY — optional, for LLM explanations
• MAFI_REDIS_ENABLED — true|false

Aegis is a multi‑agent risk system. It combines tabular, graph, behavioral, anomaly, and rule signals into a single decision. The LLM is used only for explanations, not scoring.

• If the UI shows stale data, click Reset demo data then Seed demo data.
• If reset fails, rebuild the backend container.

Proprietary / internal use.