From Siloed Algorithms to Compliance-First Agentic... | AI Research

From Siloed Algorithms to Compliance-First Agentic Platforms: A Multi-Layered Architecture for Hospital AI Systems proposes a new framework to help hospitals transition from fragmented, isolated AI tools to a unified, governed platform. The authors, Manideep Dhar, Ritwik Singh, and Sharat Chandra Kumar Manikonda, aim to address the high failure rate of healthcare AI pilots—estimated at 70-80%—by providing a blueprint that integrates governance, data privacy, and multi-agent workflows.

The Problem with Current Hospital AI

Most hospitals currently deploy AI as isolated point solutions within specific departments. This approach leads to duplicated efforts, hidden risks, and a failure to realize the full value of AI investments. The authors identify three primary causes for these failures: a lack of proper governance, fragmented data systems, and the absence of a clear integration blueprint to connect these tools across clinical, operational, and financial domains.

A Three-Layered Architecture

The proposed architecture introduces three interoperable layers designed to work with existing Hospital Information Management Systems (HIMS):

• Agent Orchestration Layer: This layer manages multi-agent workflows, allowing different AI agents to coordinate tasks across various hospital departments.

• Compliance and Policy Layer: This component centralizes "policy-as-code." It is designed to enforce global and regional standards, including HIPAA, GDPR, the EU AI Act, the DISHA Act, India's DPDP Act, and ISO/IEC security and safety standards.

• Privacy-Preserving Data Fabric: This layer integrates federated learning, differential privacy, and secure enclaves to ensure that data remains protected while being used for AI processes.

Performance and Implementation

To test the architecture, the researchers used a synthetic, structurally realistic hospital dataset and an open, ready-to-deploy prototype. The study focused on the end-to-end orchestration of triage risk prediction, workflow optimization, and compliance logging. According to the authors, the simulation showed significant reductions in task turnaround times and manual documentation effort, all while maintaining policy-guarded data access.

Strategic Value for Hospitals

The architecture is designed to be flexible, supporting on-premise, hybrid, and cloud-native deployments. By moving away from ad hoc tools, the authors suggest that hospital leaders can create a governed, globally compliant, and ROI-focused platform. This research provides a pragmatic blueprint for organizations looking to scale their AI capabilities beyond individual departmental silos.