Scaling Geo-Temporal ML - How Pokémon Go Optimizes Global Gameplay With Kubernetes and Kubeflow

Learn how Niantic's machine learning team scales geo-temporal recommendations for Pokémon Go's millions of global players using Kubernetes and Kubeflow in this 16-minute keynote presentation. Discover the unique challenges of optimizing location-based gameplay through massive geo-temporal data processing, focusing specifically on how the team revolutionized Raid Battle spawn optimization across millions of S2 cells worldwide. Explore the technical architecture behind building geo-aware recommender systems that balance player satisfaction, scalability, and privacy protection while operating at global scale. Gain insights into modeling recommendations across spatial, temporal, and difficulty dimensions, implementing geo-temporal feature engineering at scale, and applying cloud-native MLOps practices to one of the world's most popular location-based games. Understand how Kubernetes and Kubeflow enable the infrastructure needed to process and serve real-time recommendations that enhance player experiences across diverse geographic regions and time zones.