Graphene — Fund Flow Intelligence

"See through every transaction."

AI-powered Fund Flow Tracking and Fraud Detection system for the PSBs Hackathon 2026 (Union Bank of India, Problem Statement PS3).

Graphene maps end-to-end movement of funds across bank accounts, detects suspicious patterns using Graph Neural Networks and anomaly detection, and presents findings to fraud investigators through an interactive graph visualisation dashboard.

---

Features

• Graph-Based Fund Tracking — Neo4j graph database maps every money movement between accounts
• 5 AML Pattern Detection — Circular round-tripping, layering, structuring, dormant activation, profile mismatch
• GraphSAGE Neural Network — Learns structural fraud signatures from account neighbourhood context
• Isolation Forest Anomaly Detection — Unsupervised detection of novel fraud types
• SHAP Explanations — Every prediction comes with human-readable "why" explanations
• Risk Scoring Engine — Weighted ensemble: GNN (40%) + Anomaly (35%) + Rules (25%) → 0–100 score
• Interactive Dashboard — Cytoscape.js graph visualisation with real-time alerts
• FIU Report Generation — One-click PDF Suspicious Transaction Report for FIU-IND

---

Quick Start

Prerequisites

• Python 3.11+
• Neo4j Desktop or Community Edition (running on bolt://localhost:7687)

Setup

# 1. Clone and enter directory
cd graphene

# 2. Create virtual environment
python -m venv venv
venv\Scripts\activate     # Windows
# source venv/bin/activate  # macOS/Linux

# 3. Install dependencies
pip install -r requirements.txt

# 4. Copy environment file
copy .env.example .env     # Windows
# cp .env.example .env     # macOS/Linux

# 5. Start Neo4j and update .env with your credentials

Run Demo

# Generate synthetic data
python data/generate_transactions.py

# Seed Neo4j database
python data/seed_neo4j.py

# Train ML models
python ml/train.py --epochs 30 --fast

# Start API server
uvicorn api.main:app --host 0.0.0.0 --port 8000 --reload

Then open http://localhost:8000 in your browser.

---

Architecture

Layer 1 — Data Ingestion     → Synthetic transaction generator + Neo4j seeder
Layer 2 — Graph Analytics     → Neo4j Cypher pattern detection queries
Layer 3 — ML Detection        → GraphSAGE GNN + Isolation Forest + SHAP
Layer 4 — Risk Scoring        → Weighted ensemble → 0-100 risk score
Layer 5 — Investigation UI    → React + Cytoscape.js dashboard + FIU reports

---

Project Structure

graphene/
├── data/                    # Data generation and loading
│   ├── generate_transactions.py
│   ├── fraud_patterns.py
│   ├── schema.py
│   └── seed_neo4j.py
├── graph/                   # Neo4j graph operations
│   ├── neo4j_client.py
│   ├── graph_builder.py
│   ├── pattern_queries.py
│   └── graph_exporter.py
├── ml/                      # Machine learning models
│   ├── gnn_model.py
│   ├── anomaly_detector.py
│   ├── risk_scorer.py
│   ├── shap_explainer.py
│   ├── feature_engineering.py
│   └── train.py
├── api/                     # FastAPI backend
│   ├── main.py
│   ├── dependencies.py
│   └── routes/
├── frontend/                # Dashboard UI
│   └── index.html
├── requirements.txt
├── .env
└── README.md

---

Tech Stack

Layer	Technology
Graph Database	Neo4j 5.x
GNN	PyTorch + PyTorch Geometric (GraphSAGE)
Anomaly Detection	scikit-learn (Isolation Forest + LOF)
Explainability	SHAP
API	FastAPI + uvicorn
Frontend	React 18 + Cytoscape.js
PDF Reports	ReportLab

---

API Endpoints

Method	Endpoint	Description
GET	/api/health	System health check
GET	/api/alerts	List fraud alerts (filterable)
GET	/api/alerts/{id}	Alert details with SHAP
PATCH	/api/alerts/{id}	Update alert status
GET	/api/alerts/stats	Alert statistics
GET	/api/graph/{account_id}	Account subgraph
POST	/api/graph/trace	Trace fund path
GET	/api/graph/patterns	Detected AML patterns
POST	/api/reports/generate	Generate FIU STR PDF
GET	/api/reports/{id}/download	Download report

---

Risk Tiers

Score	Tier	Action
0–30	LOW (green)	Normal monitoring
31–55	MEDIUM (amber)	Enhanced due diligence
56–75	HIGH (orange)	Immediate review required
76–100	CRITICAL (red)	Freeze account, escalate to FIU

---

Team

Built for PSBs Hackathon Series 2026 — Union Bank of India

---

Graphene — "See through every transaction."