A Hybrid Cloud Framework for Secure Velocity Aggregation and Persona Enrichment

Abstract

In modern data-driven ecosystems, especially within regulated domains like healthcare and financial technology, critical data sources are often distributed across on-premises infrastructure and public cloud environments. These divisions may arise due to compliance, system design, or operational constraints. Rather than treating such separation as a limitation, this paper explores how secure collaboration between distributed systems can be leveraged to improve machine learning models through velocity aggregation, the capture and analysis of real-time transactional or behavioral patterns, and persona enrichment, the synthesis of richer user profiles from multiple secure sources. The paper focuses on scenarios where sensitive data, such as personal health records or financial transactions, must remain confidential, even during processing. To address this, we incorporate confidential computing environments (e.g., AWS Nitro Enclaves) to ensure that data processed in the cloud remains encrypted and inaccessible to unauthorized actors, including the cloud provider itself. Further, this paper proposes a hybrid cloud architecture that enables bidirectional, privacy-preserving data aggregation and federated model enhancement without exposing raw data. The framework demonstrates how secure statistical insights and model signals can be exchanged across cloud and on-prem systems, resulting in mutually improved ML model performance, while maintaining regulatory compliance and strict data confidentiality. We validate the framework using examples from healthcare and fintech, highlighting its broad applicability across any domain that handles personally identifiable information (PII).