UPI (Unified Payments Interface) transactions have become increasingly popular in India, but this growth has also led to a rise in fraudulent activities. Users often fall victim to scams during transactions, resulting in financial losses. The challenge is to detect and prevent these fraudulent transactions in real-time before they are completed.
Current solutions lack:
- Real-time fraud detection capabilities
- Multi-factor analysis of transaction patterns
- User behavior monitoring
- Immediate intervention mechanisms
FIGMENT is a multi-agent AI system designed to detect potential scams during UPI transactions in real time. Our approach combines multiple specialized AI agents that analyze different aspects of a transaction:
- Pattern Agent: Analyzes transaction messages using Machine Learning (TF-IDF + Logistic Regression)
- Network Agent: Checks receiver UPI against a crowd-reported scam database in MongoDB
- Behavior Agent: Detects anomalies in user behavior using Isolation Forest algorithms
- Biometric Agent: Monitors typing patterns (speed, corrections) for signs of coercion
The system aggregates results from all agents into an overall risk score. For high-risk transactions (>40% risk), the system requires PIN confirmation before proceeding.
Additionally, Gemini AI generates human-readable explanations for why a transaction was flagged, helping users understand the risks.
We have developed FIGMENT using two different approaches:
A full-featured web application built with React and Flask that provides a comprehensive dashboard for monitoring transactions and viewing fraud analysis.
Technology Stack:
- Frontend: React, TailwindCSS, Framer Motion, shadcn/ui
- Backend: Flask, MongoDB, scikit-learn
- AI Integration: Google Generative AI (Gemini)
A mobile-first implementation designed for on-the-go transaction monitoring and immediate fraud alerts.
For detailed instructions on setting up and running each version, please refer to the respective README files: