Evals Are a Team Sport: Building Scalable Evaluation Pipelines for Trustworthy AI

Modern AI systems fail not only because of flawed models but because evaluation is often treated as a one-time task rather than an ongoing discipline. This session addresses the challenge of scaling evaluation across teams and pipelines to ensure model reliability, fairness, and performance. Drawing from real-world experience in large-scale financial analytics, Anusha Dwivedula will examine how product and data teams can collaborate to design a continuous evaluation framework that integrates precision, recall, and drift metrics with observability, lineage, and quality controls. She will walk through how these metrics are tied to business outcomes, how anomalies are detected automatically, and how trust indicators are surfaced through dashboards and documentation. The session also introduces open-source and cloud-native tools that enable automated validation and governance in MLOps workflows. Attendees will leave with a repeatable blueprint for building scalable evaluation pipelines, establishing shared ownership between technical and product teams, and embedding trustworthy AI practices into daily delivery processes.