Inside Uber’s AI Revolution - Everything about how they use AI/ML

Michelangelo serves as Uber’s centralized machine learning platform, managing the end-to-end lifecycle of over 5,000 models that generate 28 million real-time predictions per second. To maintain scalability and developer velocity, the platform employs a four-tier prioritization system, focusing resources on business-critical core user flows like pricing and fraud detection. While historically centered on predictive machine learning, the platform is evolving to support generative AI and agentic systems through new tools like an Agent Studio and an Agent Registry. This infrastructure enables teams to build, evaluate, and deploy agents while maintaining strict internal security protocols. To accelerate innovation and keep pace with rapid industry advancements, Uber plans to open-source the Michelangelo platform, beginning with the orchestration framework Uniflow, facilitating external contributions and community-driven development.