Jose Prabhu Michael Singarayan

Beyond the Medallion: Rethinking Data Architecture for the Agentic Era

Abstract:

The medallion architecture — bronze, silver, gold — has served enterprise data teams well for over a decade. It brought order to chaos, structure to sprawl, and governance to the ungovernable. But it was designed for a world where humans posed questions and systems returned answers. That paradigm no longer holds.

Agentic AI systems don’t query data — they reason over it, act on it, and loop back to refine their own conclusions. They initiate workflows autonomously, invoke tools dynamically, evaluate outcomes probabilistically, and escalate when confidence thresholds are not met. This fundamental shift exposes the silent assumptions embedded in every lakehouse design built over the last decade: batch-first ingestion pipelines, human-in-the-loop governance models, and semantic layers that were never architected to be machine-navigable.

This keynote examines what happens when an agentic AI system is deployed on top of a conventional medallion lakehouse — and why it breaks in ways that traditional monitoring stacks cannot detect. Drawing on production deployments across retail, public sector, and life sciences environments, the talk identifies three architectural failure modes unique to agentic workloads: semantic drift under autonomous query generation, governance gaps in tool-mediated data access, and latency mismatches between agent decision cycles and pipeline refresh cadences.

The talk introduces a new architectural framework — the Reasoning-Ready Data Platform — built on four design principles: semantic observability, agent-aware access control, continuous context propagation, and feedback-loop persistence. Concrete implementation patterns are presented for organizations already invested in medallion-based lakehouses, with reference architectures applicable across both cloud-native and hybrid enterprise environments.

The medallion is not obsolete. But treating it as a destination rather than a foundation is how enterprises — and the academic frameworks that inform them — will fall behind the agentic curve.

Profile:

Jose Prabhu Michael is a seasoned AI Data Architect and IEEE Member with over 20 years of experience in designing, implementing, and governing large-scale enterprise data platforms across retail, finance, and public sector environments. He has a strong background in AI-native architecture, semantic layer design, and data platform modernization, with proven expertise in leading complex data transformation initiatives across global enterprise environments.

Jose specializes in Microsoft Fabric medallion lakehouse architecture, generative AI integration, and agentic BI, along with implementing data governance frameworks to optimize enterprise decision-making. He has demonstrated leadership in aligning data strategy with business objectives, managing cross-functional stakeholders, and delivering scalable, AI-ready platforms that bridge the gap between raw data and autonomous intelligence.

In addition to technical leadership, Jose has a strong research and academic profile, serving on the Technical Program Committee for multiple international IEEE conferences and having peer-reviewed 50-plus research papers spanning data engineering, machine learning, and AI governance. He is recognized for driving innovation in enterprise AI, advancing semantic observability practices, and mentoring organizations to adopt agentic AI architectures — making him a valued contributor to enterprise digital transformation and AI readiness initiatives.