ML Platform Journey
summary
SurveyMonkey describes moving from ad hoc notebooks to a standardized ML platform that serves 100 million predictions/day, cuts model deployment time from months to weeks, and halves training/deployment cost. Engineering builds a custom gateway that moves feature extraction to the edge, supports async event-driven predictions and AB/shadow testing, and reduces model service nodes from eight to two, yielding an average 126 ms saving and better resource use. Focus areas: hybrid cloud, central feature store in progress.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 2:30Journey so far
- 5:58ML Platform
- 7:49Old Prediction Flow
- 12:50Prediction Flow Overview
- 14:36Benefits of Prediction Flow
- 15:36Model Management
- 19:27Recap
- 21:23Questions
related talks
30mML Platform for Insurance Conglomerate
09mPlatform Engineering for Data Science at Scale
23mML Platform Challenges & Applications at Freshworks
31mBuilding the Business case for ML Platforms
19mSustaining ML in the enterprise
26mManaging ML Models @ Scale
44mML in Production
28mBuilding Developer-Centric ML Inference Platforms