🎥 Recorded live at the MLOps World | GenAI Summit 2025 — Austin, TX (October 9, 2025) Session Title: Insights and Epic Fails from 5 Years of Building ML Platforms Speaker: Eric Riddoch, Director of ML Platform, Pattern AI Talk Track: ML Collaboration in Large Organizations Abstract: What does it really take to build an ML platform that scales—and what can go wrong along the way? In this candid and practical session, Eric Riddoch, Director of ML Platform at Pattern AI, reflects on five years of designing and operating ML platforms serving over 14 million YouTubers and the largest third-party seller on Amazon. Through lessons learned across three platform builds, Eric shares the architectures, strategies, and “epic fails” that shaped his approach to MLOps at scale. He dives into tool selection, drift detection myths, offline vs. online inference tradeoffs, and the real-world balance between data science autonomy and engineering rigor. This talk offers a refreshing, no-hype perspective on how to design ML platforms that achieve adoption, stability, and real business value. What you’ll learn: • Principles tools: All MLOps tools map to ~9 “jobs to be done” • Why drift monitoring is overrated—and data quality issues are the real culprits • When to use offline inference instead of serving every model as an endpoint • How data lineage prevents target leakage and systemic errors • Why medium-sized data tools often outperform over-engineered stacks • When non-hyperscaler cloud GPUs can outperform on-prem alternatives.