README.md

metadata

title: Enhanced Fraud Detector Agent
emoji: 🔍
colorFrom: red
colorTo: red
sdk: gradio
sdk_version: 5.42.0
app_file: app.py
pinned: false
license: apache-2.0
python_version: 3.12
suggested_hardware: cpu-upgrade
tags:
  - fraud-detection
  - langchain
  - machine-learning
  - finance
  - security
short_description: AI-powered fraud detection with multi-model & RAG.

🔍 Enhanced Fraud Detection Agent

A production-ready fraud detection system built with LangChain, Hugging Face models, and advanced ML techniques.

Features

• Multi-Model Analysis: Uses Ministral-8B for analysis and Qwen2.5-Coder for coding tasks
• Velocity Detection: Identifies suspicious transaction patterns in time windows
• Amount Anomaly Detection: Statistical analysis for unusual transaction amounts
• RAG System: Historical pattern matching using vector embeddings
• User Authentication: Secure login system with per-user data isolation
• Report Generation: Professional PDF and CSV reports
• Email Alerts: Automated notifications for suspicious activities

Quick Start

1. Upload Transaction Data: CSV file with columns: timestamp, amount, description, merchant (optional)
2. Login/Register: Create an account or login to access the system
3. Analyze: Upload your CSV and optionally provide a RAG query
4. Review Results: Get detailed analysis, flagged transactions, and downloadable reports

Sample CSV Format

timestamp,amount,description,merchant,category
2024-01-15 10:30:00,25.99,Coffee Shop Purchase,Starbucks,Food
2024-01-15 10:31:00,1500.00,Gift Card Purchase,Amazon,Shopping
2024-01-15 10:32:00,2000.00,Wire Transfer,Bank,Transfer

Configuration

The system uses the following models:

• Analysis Model: mistralai/Ministral-8B-Instruct-2410
• Code Model: Qwen/Qwen2.5-Coder-32B-Instruct
• Embeddings: sentence-transformers/all-MiniLM-L6-v2

Security Features

• Password hashing with SHA-256
• Per-user data isolation
• Session management
• SQL injection protection through parameterized queries

Technical Stack

• LangChain: Agent orchestration and tool management
• Hugging Face: Model inference and embeddings
• FAISS: Vector storage for RAG
• Gradio: Web interface
• SQLite: User and transaction data storage
• ReportLab: PDF report generation

License

Apache 2.0