Designing Data Driven Projects: Beyond ML Experimentation
summary
A grocery recommendation prototype goes live, shows value, then repeatedly fails in production: a static product file expired, ephemeral EMR infrastructure wasn’t cleaned up, and only an executable object lives in AWS so no one can recover the original code or data. Solution checklist: start with observability (structured logs and alerts), add an evaluation loop that captures ground truth, implement unit and data tests, build a reproducible training pipeline and enforce model traceability with versioned data, models and retriable infra.
AI-written summary · about
chapters (5)
- 0:00Start
- 3:52Chapters
- 20:09Observability Infrastructure
- 21:15Evaluation First Approach
- 33:39ML Product Evolution
related talks
36mApplied Machine Learning Playbook: From Idea to Production
48mPrinciples and Practices of Building Machine Learning Systems
27mBuild Your Own ML Data Feedback Loop
38mData Secrets from a Platform Engineer
39mDoing Data: The Critical Process of Data Preparation (...)
32mTips and tricks: data science prototype into production
30mFrom Black Box to a Known Quantity: How to Build Predictable, Reliable ML-based
01mintro video