AI-Powered Anomaly Detection In Fintech: Bridging Devops With Large Language Models For Scalable Fraud Prevention

Abstract

The financial technology sector is experiencing rising fraud risks as digital systems grow more interconnected. This study proposes a data-driven decision-support framework that combines Large Language Models (LLMs) with anomaly detection methods to strengthen fraud prevention. Unlike static rule-based systems, our approach applies network analysis and optimization techniques to capture contextual anomalies across diverse financial transactions. The framework uses adaptive data preprocessing, scalable decision pipelines, and contextual risk scoring to improve real-time detection while reducing false positives. Through case studies in transaction monitoring, identity verification, and payment processing, we observed 30–45% higher detection accuracy and significant reductions in false alerts. These findings suggest that LLMs can serve as a practical decision-support tool for managing financial risk. We also discuss deployment challenges and outline future directions such as federated learning, explainable AI, and distributed optimization, which can enhance scalability, transparency, and resilience against evolving fraud tactics.