Market Manipulation Detection & Insider Trading Prevention System

AI-driven financial surveillance platform that detects insider trading and market manipulation (pump-and-dump, spoofing, layering) using a multi-modal ensemble of Isolation Forest, LSTM Autoencoders, Graph Neural Networks, and FinBERT sentiment analysis on real-time NSE stock data.

Setup Instructions

Prerequisites

• Python 3.10+
• Node.js 18+

Backend Setup

1. cd backend
2. Create virtual environment: python -m venv venv
3. Activate it:
   • Windows: venv\Scripts\Activate.ps1
   • Mac/Linux: source venv/bin/activate
4. Install dependencies: pip install -r requirements.txt
5. Copy .env.example to .env and fill in your own credentials (Gmail App Password, Twelve Data API key)
6. Run: uvicorn app.main:app --reload --port 8000

Frontend Setup

1. cd frontend
2. Install dependencies: npm install
3. Run: npm run dev
4. Open http://localhost:5173

Notes

• Postgres/Neo4j are optional — the app automatically falls back to SQLite / an in-memory trader graph if they're not running locally, so it works out of the box without Docker.
• On first run, inject a few scenarios and click "Fast Forward" a couple times to warm up the ML models (LSTM/GNN need some baseline data to train on) before demoing.
• Never commit your .env file — it's already excluded via .gitignore.

Features

• Real-time price simulation blended with live Twelve Data API feed
• Multi-model risk scoring: Isolation Forest, LSTM Autoencoder, Graph Neural Network, FinBERT sentiment
• Supervised RandomForest classifier trained on confirmed alert patterns
• Automated email reporting and complaint filing via Gmail SMTP
• Explainability layer (why/immediate issue/recommended action per alert, model contribution breakdown)
• Trader network visualization and wash-trading detection
• Live news feed integration via Google News RSS
• Dashboard tabs: Model Comparison, Evaluation Metrics, Trader Reputation, News Feed