Leveraging Large Language Models in a Secure AWS Cloud Framework for Federated Learning–Driven Predictive Analytics across Financial and Healthcare Domains

Abstract

The increasing adoption of data-driven intelligence in financial and healthcare systems raises critical challenges related to data privacy, security, and regulatory compliance. Centralized machine learning approaches often fail to address these concerns due to the sensitive nature of financial records and electronic health data. This paper proposes a secure AWS cloud-based framework that leverages Large Language Models (LLMs) and federated learning to enable predictive analytics across distributed financial and healthcare domains. The proposed architecture integrates privacy-preserving federated learning mechanisms with LLM-driven analytics to support real-time insights without exposing raw data to centralized repositories. AWS-native security services, including identity and access management, encryption, secure data storage, and monitoring, are employed to ensure compliance with industry regulations such as HIPAA and PCI-DSS. The framework supports scalable model training, secure inference, and adaptive risk-aware analytics across heterogeneous data sources. Experimental analysis demonstrates that the proposed approach enhances predictive accuracy while significantly reducing data leakage risks and communication overhead. This research highlights the potential of combining LLMs and federated learning within a secure cloud environment to deliver trustworthy, real-time predictive analytics for mission-critical financial and healthcare applications.