Michelangelo ML Platform at Uber: Past, Present and Future
summary
Michelangelo begins in 2016 and now combines a code-first engine (Canvas, launched 2020) and a unified UI (MS Studio) with a Kubernetes-style unified API (Kubernetes CRD pattern) that version-controls specs/status via Git and auto-generates code for reviews. Built components include open‑sourced Horovod for distributed TF/PyTorch training and the Palette feature store (point‑in‑time joins, ~10M QPS). Outcome: deep learning adoption expands (most TR1 models migrated), developer productivity and UI fragmentation noticeably shrink, and plans add an LLM gateway, fine‑tuning and serving optimizations.
AI-written summary · about
chapters (6)
- 0:00Start
- 7:44Project Canvas Streamlined E2E developer experience with first-class DL support
- 9:39Developing Canvas Applications
- 13:50Unified API Framework
- 14:56Case Study: Pipeline Management
- 18:00Summary
related talks
25mAccelerating ML at Uber with the Palette Feature Store
17mKeynote: Machine Learning & A.I. At Uber
31mUse-Case Driven ML Platforms
33mScaling PyTorch Training with Ray Data on Uber's Michelangelo AI Platform
33mUber's GenAI Leap: Batch Predictions Using Ray and vLLM
26mUber’s Intelligent Insights Assistant
42mMLOps 2025: A Journey into the Past and the Future
43mMachine Learning Engineering principles with Python and MLFlow