Jamie Petherbridge-Conroy

Imagine Netflix launching News, or Spotify launching a learning platform. Now imagine trying to personalise across all of that, plus kids, live sport, and radio/audio with a consistent user experience and public‑service editorial standards. Discoverability teams at the BBC serve personalised content to over a billion requests a month across domains ranging from Video on Demand to Radio to News and Sport. To be able to do this reliably at scale, MLOps becomes fundamental. In this session, discover how the BBC builds and improves numerous Machine Learning and AI models at scale and pace, across multiple domains, whilst delivering value for license fee payers. This session introduces how we split end-to-end reliability into 6 key areas to structure our MLOps work and enable Data Scientists to keep pace with larger, better funded competitors. Takeaways: - A 6-part ML reliability framework you can apply to recommender/discovery systems in complex organisations - Concrete examples of what worked and what didn’t when operationalising MLOps across multiple BBC domains - How reliability investment reduced firefighting and unlocked higher experiment throughput and faster model iteration