Abstract

This research introduces a unified system that integrates large language models (LLMs) and cloud security mechanisms within web systems to improve financial fraud detection and ETL-based data integration. The proliferation of digital financial transactions and complex data pipelines necessitates intelligent approaches that can both analyze textual/transactional anomalies and secure sensitive data flows. We propose a hybrid architecture that combines advanced LLMs trained on financial datasets with cloud security services, enabling real-time fraud identification through natural language and pattern recognition, while simultaneously safeguarding data during extraction, transformation, and loading (ETL) processes. This model leverages scalable cloud platforms (e.g., serverless functions, container orchestration) and secure key management to ensure compliance with privacy standards and enhance resilience against threats. Experimental evaluations on benchmark datasets demonstrate significant improvements in fraud detection accuracy and latency compared to traditional rule-based systems. The unified model also achieves robust data integration performance with low error rates and high throughput. Findings suggest that incorporating AI LLM capabilities within secure cloud infrastructures can empower organizations to detect sophisticated threats and maintain integrity across distributed web systems. We discuss design choices, implementation challenges, and implications for future intelligent financial systems.