← Back to Projects Finance 
Real-time Fraud Detection System
Production-grade ML pipeline for credit card fraud detection with real-time streaming inference, SHAP explainability, A/B testing, and a Streamlit monitoring dashboard.
Overview
Production-grade, end-to-end ML pipeline for credit card fraud detection with streaming data processing, real-time feature engineering, and model inference.
Architecture
- Streaming transaction processing with async queues
- Real-time feature engineering pipeline
- XGBoost model with SMOTE for class imbalance
- SHAP-based explainability for every prediction
- FastAPI serving + Streamlit multi-page dashboard
Key Features
- Real-time fraud scoring with sub-100ms latency
- A/B testing framework for model comparison
- Performance monitoring dashboard with drift detection
- Feature importance visualization with SHAP