Uber’s Intelligent Insights Assistant
summary
Uber reports platforms that turn internal users into analysts: a forecasting platform that auto-preprocesses, backtests and selects models (classical, ML, hybrid) and serves point forecasts plus prediction intervals via Michaelangelo, a real-time anomaly stack and an experimentation system, and conversational features like one-click chat and voice dispatch that cut handle time by 8.8%. Kepler GL is open sourced with ~35,000 external users and 5,500 stars. A proof-of-concept insights assistant scans data and surfaces novel findings; a 500M time series outage detector targets 99.99% availability.
AI-written summary · about
chapters (19)
- 0:00Intro
- 0:11Data Science is key to Uber's future
- 2:16Data Science Project Lifecycle
- 6:08Multiplier Effect through Specialists Team
- 6:41Forecasting Platform
- 7:433 of Many Forecasting Use Cases at Uber
- 11:36Data Exploration and Preprocessing
- 12:08Forecasting Methodologies
- 13:39Backtesting under the Hood
- 14:52Automated Backtesting at Scale
- 15:38Blackbox Feature Extraction and Event Forecasting
- 17:34At Scale Integration and Quality Assurance
- 18:48Overview of One-Click Chat
- 19:24Intent Detection Challenges
- 20:183 of Many NL / Conversational Al Use Cases at Uber
- 21:59NL / Conversational Al Platform Components
- 22:40Exploratory Analysis at a Push-of-a-button
- 23:30New Business Insights Suggestions
- 24:203 Phases of Commoditization
related talks
17mKeynote: Machine Learning & A.I. At Uber
18mAI Platform Q&A
25mAccelerating ML at Uber with the Palette Feature Store
35mUber's Journey in Building Multi-Modal Foundation Models
13mPyTorch at Uber
33mUber's GenAI Leap: Batch Predictions Using Ray and vLLM
19mMichelangelo ML Platform at Uber: Past, Present and Future
31mUse-Case Driven ML Platforms