An SRE Approach to Monitoring ML in Production

This 39-minute conference talk from SREcon25 Americas explores how Site Reliability Engineering (SRE) principles can be applied to monitoring Machine Learning systems in production environments. Presented by Daria Barteneva from Microsoft Azure, discover the challenges of operationalizing ML within large distributed systems and the expertise gap between ML development and production reliability. Learn how to decompose complex ML systems into observable components, understand why traditional observability practices fall short for ML workloads, and explore mechanisms for monitoring end-to-end system reliability and quality. Gain practical insights for SREs who are or will soon be responsible for serving ML models at scale in production environments.