Project SENTINEL | End-to-End Proposed Solution | PayVista AI Fraud Detection System

🚀 Project SENTINEL – End-to-End Proposed Solution In this video, I explain my proposed end-to-end solution for Project SENTINEL, an AI-powered Fraud Intelligence System for PayVista. The objective is to move beyond a traditional rules-based fraud detection system and build SENTINEL-LM, capable of understanding transaction context, customer behaviour and fraud patterns to support real-time fraud decisions. 🔹 Topics covered in this video: • PayVista historical transaction data • Data cleaning and PII removal • Feature engineering for fraud detection • XGBoost baseline model • SENTINEL-LM development • Continued Pre-Training • Supervised Fine-Tuning (SFT) • Direct Preference Optimization (DPO) • Distillation and Quantisation • Real-time Fraud Model API • Redis Feature Store • Fraud probability and risk scoring • Rules Engine • 🟢 ALLOW | 🟡 REVIEW | 🔴 BLOCK decisions • AWS deployment • Shadow-mode testing and gradual rollout • Monitoring, feedback and model retraining The complete solution starts with PayVista's historical fraud data and ends with a production-ready fraud intelligence system capable of making fast, explainable and secure fraud decisions. This video focuses on solution thinking and explains how the complete system can be designed, trained, deployed, monitored and continuously improved. #ProjectSENTINEL #PayVista #FraudDetection #ArtificialIntelligence #LLM #MachineLearning #GenerativeAI #FraudPrevention #AIEngineering #AWS #XGBoost #FastAPI #Redis #MLOps