Vandana Kollati

Governance‑First Evaluation of Agentic AI Systems for Regulated Enterprise Data Intelligence

Abstract:

Agentic Artificial Intelligence (AI) systems capable of autonomous planning, tool invocation, and workflow orchestration are increasingly adopted in regulated enterprise domains such as healthcare, finance, and public administration. While these systems offer significant operational benefits, their autonomy introduces critical governance and security risks, including unauthorized data access, regulatory non‑compliance, and orchestration‑layer attacks under frameworks such as HIPAA, GDPR, and the EU AI Act. Existing comparative studies largely emphasize functional performance and neglect real‑world governance requirements.

This paper presents a governance‑first comparative evaluation of three agentic AI platforms: Microsoft Fabric Copilot, Google Gemini agents, and the Meta LLaMA stack. Using the identified New York SPARCS hospital discharge dataset and the UNCTAD World Government Expenditure dataset, we executed regulated enterprise workflows with uniform privacy, access control, and auditing constraints. Results show that Microsoft Fabric Copilot achieves the highest governance maturity with complete auditability and zero unauthorized access incidents, while Gemini demonstrates strong scalability and latency. The open‑source LLaMA implementation exhibits higher compliance violations requiring additional hardening. We further introduce a four‑layer governance and cyber‑resilience framework that reduces compliance violations by 70% and unauthorized access attempts by 95%.

Profile:

Vandana Kollati is a technical data architect, data engineering, and enterprise data governance professional with extensive experience in designing secure, compliant, and scalable analytics platforms. Her work focuses on agentic AI systems, data quality frameworks, metadata governance, and regulatory‑aligned data intelligence architectures. She has led and contributed to multiple enterprise initiatives involving cloud‑native analytics, automated data quality, and AI governance in highly regulated environments.

Her technical expertise includes the design of end‑to‑end enterprise data architectures encompassing data ingestion, lakehouse and warehouse patterns, metadata‑driven governance, access control models, and audit‑ready analytics workflows. She has architected resilient data platforms that integrate data quality enforcement, lineage, and policy‑based controls into AI‑enabled analytics systems.

Vandana is an active member of professional data engineering, artificial intelligence, and data governance communities, where she engages in knowledge sharing, technical discussions, and best‑practice development. She has authored and co‑authored technical articles and research‑oriented publications in the areas of enterprise analytics, AI governance, data architecture, and data quality. She also serves as a reviewer and judge for technical papers, solution evaluations, and innovation initiatives, contributing to the assessment of emerging technologies and responsible AI practices.