From POC to production: Strategies for achieving machine learning at scale
summary
Seven concrete steps for moving from POC to production: establish culture and goals, set up cross‑functional teams and a COE, secure production data with either a sandbox vs platform approach, and automate pipelines as repeatable MLOps. Proofs of concept typically run 2–6 month POC timelines; production patterns include AB testing and shadow deploys for safe rollout. Amazon SageMaker is recommended — Intuit cuts ML dev time from six months to one week, a 90% development time reduction.
AI-written summary · about
chapters (18)
- 0:00Introduction
- 0:16Agenda
- 1:40Achieving ML at scale
- 6:25Establishing team norms expectations
- 10:00Team sports analogy
- 10:22Team roles
- 14:22How are models trained
- 16:28Data security
- 18:48Establish a data platform
- 20:19Planning a proof of concept
- 23:41Automation
- 24:51Building repeatable pipelines
- 25:52Machine learning center of excellence
- 27:44The journey is done
- 29:00Amazon Sagemaker
- 30:58GE Healthcare
- 32:10Resources
- 33:01AWS Partner Network
related talks
10mFrom Model to Production: Building Scalable MLOps Pipelines
13mMLOps: From AI Research to Production-Ready Systems
32mAutomating Production Production Level Machine Learning on AWS
45mStart your AI/ML Projects Right
43mDesigning and Building Complex Machine Learning Projects
47mMachine Learning in Production: An Intro to MLOps.
44mPragmatic Machine Learning in the Cloud
01h 05mMachine Learning Models and Productionization