How did we build an AI/ML Platform at DoorDash from the ground up
summary
DoorDash builds an ML platform starting in 2019 by productizing ETA prediction as the hero use case, then generalizing to search, recommendations, fraud and contactless delivery. Architecture begins as a centralized service and evolves to an isolated multi‑tenant model. Engineering choices include Redis for a low‑latency feature store then migrating parts to CockroachDB; optimizations cut cost by ~33% and CPU by 38%, and the platform now serves 5 million predictions per second.
AI-written summary · about
chapters (7)
- 0:00Introduction and context
- 1:31Overview of AIML at DoorDash
- 4:40Lesson 1: Know your game
- 9:29Lesson 2: Dream big, start small
- 13:23Lesson 3: Iterate constantly
- 18:14Generative AI and future steps
- 19:41Q&A and closing remarks
related talks
31mUse-Case Driven ML Platforms
32mTeam Teardown: Evolving an ML Platform
20mReal-time Machine Learning Inference Platform
33mExecutive Summit Roundtable Discussion
29mBuilding ML Platform Capabilities at Global Scale
17mLaying the foundation for deep learning
29mBuilding an ML Platform from Scratch
09mPlatform Engineering for Data Science at Scale