Somesh Nagalla

From Models to Agents: Rebuilding the Data Platform for Autonomous AI

Abstract:

Abstract For a decade, applied AI ran on a simple contract: train a model, ship it, call it for a prediction — and the data platform's job was to feed it features and store its outputs. Agentic AI breaks that contract. Systems no longer just predict; they plan, act, call tools, and loop toward a goal — invalidating the assumptions our infrastructure was built on: batch not real-time, read not read-write, one call not thousands, deterministic not stochastic. This keynote looks at applied intelligence from the layer beneath the algorithms and asks what must change for autonomous AI to work at enterprise scale — from model-serving platforms to agent-native ones: real-time context and memory, a governed action layer, guardrails, and observability for non-deterministic behavior. It closes with open problems for this community.

Profile:

Somesh Nagalla is a Senior Software Engineer on the Enterprise Data Platform team at 7-Eleven, Inc., where he builds the large-scale data and machine-learning infrastructure behind enterprise AI — Spark/Databricks pipelines, cloud-native platforms on Kubernetes, and CI/CD systems supporting millions of daily transactions. Across more than a decade in software and platform engineering, he has held roles at Verizon, and holds an MS in Technology Management from the University of Bridgeport. His research spans applied AI, cloud computing, big data, and scalable ML systems, with a growing publication record across IEEE and Springer/Scopus venues and contributions merged into widely used open-source projects including Apache Iceberg and sktime, as well as NASA and CISA repositories. In 2026 he serves as an invited and keynote speaker at several international conferences, focusing on the infrastructure that makes applied AI reliable, efficient, and trustworthy at scale.