Diana Omuoyo

AI systems have introduced new operational challenges: unpredictable model behaviour, data drift, prompt variability, escalating infrastructure costs, regulatory obligations, and growing concerns around data sovereignty and security. This session explores how observability practices need to evolve for operating AI at scale. Drawing on real-world lessons from modern observability practices, we will examine the five critical pillars of AI observability: Infrastructure Health, Model Performance, Data Quality, Cost & Security. Attendees will learn how to build end-to-end visibility across the entire AI lifecycle. The presentation will demonstrate how telemetry, tracing, logging, and AI-specific signals can be combined to detect model degradation, identify security risks, validate compliance requirements, and provide actionable insights for platform and operations teams. Whether you are responsible for platform engineering, security operations, or AI infrastructure, this talk will provide a practical framework for implementing observability strategies that enable reliable, effecient and secure AI systems.