Overview
RiskGuard ML is a comprehensive fraud detection system that combines multiple machine learning approaches to identify fraudulent transactions in real-time. The system uses ensemble methods with XGBoost and Random Forest for high accuracy, while PyTorch-based neural networks handle complex pattern recognition.
Key Highlights
- Built supervised and semi-supervised fraud detection models using XGBoost, Random Forest, and PyTorch on 5M+ transactional records
- Implemented cross-validation, ROC-AUC optimization, and cost-sensitive threshold tuning to improve fraud detection recall by ~19%
- Deployed real-time risk scoring APIs using Flask, Redis caching, and Kubernetes with monitoring and drift detection pipelines
Technologies Used
Large Language Models (GPT-based, OpenAI API, HuggingFace)Prompt EngineeringTool and Function CallingRetrieval-Augmented Generation (RAG)Embedding ModelsSemantic RetrievalAutonomous ReasoningDecision Orchestration