Real-Time Fraud Detection Pipeline

End-to-end MLOps streaming pipeline with live XGBoost inference & monitoring

Production-style real-time credit card fraud detection system. Transactions are streamed via Redpanda, processed with feature engineering, scored instantly by a trained XGBoost model, stored in PostgreSQL, and visualized live on Grafana dashboards showing fraud volume and financial impact. Fully containerized with Docker Compose.

• › Producer → Redpanda → Consumer architecture for live streaming
• › Real-time XGBoost inference on 30+ features per transaction
• › Live Grafana monitoring of fraud metrics and potential losses
• › End-to-end MLOps: ingestion, processing, prediction, storage & observability
• › Containerized setup with one-command infrastructure spin-up