Real-time Machine Learning Inference Platform
summary
Zomato’s platform for real-time ML inference, built to support use cases like kitchen preparation time, rider assignment time, and personalised restaurant ranking, is described including the technologies used for serving models and the deployment workflow. The talk covers how the deployment process was reworked to reduce time-to-production from one to two months down to under a week and the outcomes achieved for model deployment.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:30What is Machine Learning ?
- 2:36ML @Zomato
- 4:25Challenges
- 5:50Requirements
- 7:19KPT Prediction
- 8:12Proposed Architecture
- 10:51Real time feature Computation by Apache Flink
- 12:04Real Time Feature Store
- 13:38Flink - Feature Store
- 13:59Static Feature Store
- 16:18Deploying models through AWS Sagemaker
- 17:24Enriching using ML Gateway
- 18:00KPT Plan
- 18:13Final Architecture
- 19:04Future Work
related talks
30mReal-Time ML services: From ideation to production
30mHow did we build an AI/ML Platform at DoorDash from the ground up
25mMachine Learning is Going Real-Time
32mLiveMLP: ML Platform for Assisting Contact Center Agents in Real-Time
24mBuilding an ML Application Platform from the Ground Up
04mInterview
17mKeynote: Machine Learning & A.I. At Uber
53mMachine Learning Platforms