Use-Case Driven ML Platforms
summary
A platform team of 100 builds higher-level ML services on top of Michelangelo and reports concrete wins: push-button forecasting integrated into the BI dashboard that scans many algorithms and returns forecasts without user tuning, an NLP ticket-assistant that improves customer care responses, and real-time anomaly detection on hundreds of millions of signals that achieves very low latency and holds patents. Engineering choices include microservices in Go/Java, MySQL on-prem, and pi-ml model deployments enabling self-service productionization.
AI-written summary · about
chapters (19)
- 0:00Intro
- 0:45Team and Charter
- 2:33conversational AI
- 3:54when to platform
- 6:00relationship with customers
- 6:43methodology
- 8:52relationship with Uber
- 11:49dynamic relationship
- 13:14methodologies and workflows
- 13:35technical challenges
- 15:47building from the ground up
- 16:22technology stack
- 17:26open source
- 19:15systems engineering
- 21:10team structure
- 22:46real time anomaly detection
- 25:21Agile methodologies
- 26:58How to grow your team
- 28:50Future of data science platforms
related talks
28mBuilding Developer-Centric ML Inference Platforms
31mBuilding the Business case for ML Platforms
19mMichelangelo ML Platform at Uber: Past, Present and Future
30mML Platform for Insurance Conglomerate
09mPlatform Engineering for Data Science at Scale
32mMaking Models Work: How an optimized ML lifecycle can drive better ROI
17mKeynote: Machine Learning & A.I. At Uber
38mFantastic ML apps and how to build them