Satark: Collaborative Fintech Security Platform (Phase 1)

Satark (Hindi for "Alert/Cautious") is a deterministic Fintech and Graph-Networking engine designed to identify anomalous clusters, multi-hop financial fraud, and money laundering (like Smurfing rings) without sharing raw customer data.

Instead of dealing with computationally heavy and error-prone ML heuristics like OCR, Satark operates directly on the financial schema level, catching coordinated fraud networks before capital is fully siphoned.

The Architecture & Workflow

SATARK operates on a Federated Intelligence Model. Instead of sharing raw customer data, banks connect their localized nodes to a central Neural Kernel that analyzes cryptographic identifiers.

Step-by-Step System Flow:

1. Interactive Entry: Institutional portals featuring cursor-tracking spotlight heuristics for forensic focus.
2. Institutional Ingestion: High-throughput transactions (5,000+ nodes) are ingested via a FastAPI gateway.
3. Distributed Analysis: Celery workers execute Tarjan's SCC and Isolation Forest algorithms in parallel.
4. Cinematic Visualization: Live transaction topology is rendered in a 3D interactive galaxy for analysts, utilizing a nested spherical physics lock.
5. Cross-Bank Verification: Interactive simulation allowing analysts to verify anomalous nodes with external institutions in real-time.

The Stack

• Frontend: React 18 (TypeScript), Vite, Tailwind CSS, Framer Motion, react-force-graph-3d, Three.js.
• Backend: FastAPI, SQLite (Relational), Neo4j (Graph), networkx, scikit-learn.
• Distributed Queue: Celery (gevent pool for Windows), Redis.
• Data Integrity: HMAC-SHA256 Payload Validation, Zero-PII Cryptographic Masking.

How to run this locally

Since we are running an asynchronous queue, getting this running requires a couple of specific terminal windows.

Terminal 1: The Backend (FastAPI)

python -m venv venv
.\venv\Scripts\activate  # Windows
source venv/bin/activate # Mac/Linux

pip install -r requirements.txt
python generate_threat_network.py # Seeds the Neo4j and SQLite DB with 5,000+ nodes and fraud patterns
uvicorn main:app --reload

The server will boot on http://localhost:8000

Terminal 2: The Task Queue (Celery)

.\venv\Scripts\activate
# For Windows, use solo pool or gevent
celery -A tasks worker --loglevel=info --pool=solo

Terminal 3: The Frontend (Vite/React)

cd frontend
npm install
npm run dev

The React portal will boot on http://localhost:5173

Upcoming: Phase 2

In the Phase 2 update, we will incorporate federated learning across multiple banks to preserve total graph anonymity and implement Private Set Intersection (PSI) for secure cross-bank matching.

Live Data Generator

The generate_threat_network.py script automatically builds the threat databases. It intentionally injects complex fraud patterns (Smurfing, Cycles, Velocity Bursts) so the Admin Dashboard and 3D Visualizer populate beautifully on boot.