I speak about the parts of machine learning that happen after the notebook closes: keeping ranking systems reliable, fast, and honest under real production load.

• When Your Model Meets Reality: Designing ML Ranking Systems That Survive Billions of Daily Requests — ODSC AI West (virtual talk). On inference pipelines under strict latency budgets, feature systems that stay reliable when upstream data degrades, graceful degradation, and the feedback loop between production performance and retraining.