Managing ML Models @ Scale
summary
Intuit’s ML Platform claims 200% increase in model publishing velocity by combining GitOps, Argo Workflows, Kubernetes and AWS SageMaker with a feature store that separates online/offline feature backs and central metadata search. It runs Beam streaming and Spark batch feature pipelines, codifies training via Argo workflows, offers perf tests that recommend instance sizes, attaches billing tags per model for cost attribution, and enforces container scanning, VPC-restricted endpoints and automated compliance controls.
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:01ML driven experiences
- 2:18"Hidden Debt" of ML
- 2:51How can ML Practitioners focus on their craft?
- 3:16Principles of Intuit ML Platform
- 3:56Generic Model Lifecycle
- 5:02Technologies in use
- 5:37Feature Store
- 7:27Feature Processing
- 8:42Model Training
- 12:56Automate the simple things
- 15:00With great speed, comes cost
- 16:52Cost transparency
- 18:08What can go wrong?
- 19:21Security
- 20:56Compliance
- 22:58Learnings
related talks
27mScaling ML to the next level: Hosting thousands of models on SageMaker
30mML Platform for Insurance Conglomerate
28mScaling MLOps on Kubernetes with Amazon SageMaker Operators
36mEnterprise-grade Full Stack ML Platform
30mCost Transparency and Model ROI
32mAutomating Production Production Level Machine Learning on AWS
32mMaking Models Work: How an optimized ML lifecycle can drive better ROI
28mBuilding Developer-Centric ML Inference Platforms