Scaling Instacart Fulfillment ML on Ray

Discover how Instacart scaled its fulfillment machine learning capabilities using Ray in this Ray Summit 2022 conference talk. Learn about the challenges of predicting regional-level fulfillment metrics across 2,000 unique zones and how Instacart transitioned from a Celery-based queueing system on AWS ECS to Ray distributed frameworks. Explore the benefits of this transition, including increased parallelization, cost reduction, and significant improvements in training time. Gain insights into Instacart's serverless architecture, remote code execution, workspace isolation, and CPU utilization optimization. Examine the integration of Ray with Instacart's ML Ops and ML Platform, and understand the next steps in their machine learning journey.

Intro
What is Instacart
Examples of fulfillment ML
Previous solution
Pseudocode
New solution
Serverless architecture
Remote code
Work workspace isolation
Better CPU utilization
Early testing results
Integration with ML Ops
Instacart ML Platform
ML Launcher Integration
Next Steps