Pravin Khandke

The Trust Gap: Architecting Autonomous AI Systems for Real-World Accountability

Abstract - Autonomous AI agents are being deployed across industries at an accelerating pace, yet the gap between their promised capabilities and their real-world reliability continues to widen. This keynote examines the trust gap from three angles: the architecture of agent systems that break in production, the verification methods needed to hold them accountable, and the sustainability costs that most deployments ignore. Drawing on practical experience with multi-agent orchestration, human-in-the-loop verification, and carbon-aware AI routing, it introduces a layered trust framework spanning design-time validation, runtime monitoring, and post-hoc audit. The talk covers agent failure modes in deployed systems, patterns for building verifiable agent pipelines, the role of physical provenance in authenticity verification, and sustainability-aware architectures that track their own environmental footprint. It concludes with emerging directions in self-auditing agents, trust-as-a-service infrastructure, and the regulatory implications of autonomous decision systems.

Brief Profile - Pravin Khandke is an IEEE Senior Member, SCRS Fellow, and SAS Eminent Fellow with 26 years of experience architecting large-scale AI and distributed systems across retail, automotive, hospitality, and financial services. He directed the nine-year modernization of a platform processing millions of transactions daily across North America's largest automotive marketplace and engineered adversarial-resilient automation that monitored over 5,000 hotel booking sites at production scale. His systems have processed data across thousands of retail nodes and shaped canonical integration standards adopted across independent platforms. He has completed over 60 peer reviews across 17 IEEE and international venues and serves on 4 technical program committees. His current work bridges production experience with scholarly research in multi-agent architectures, carbon-aware AI orchestration, and referential authenticity verification. An active contributor to the IEEE and SCRS communities, his research focuses on building AI systems that can be trusted not just in theory but in production.