Principles and Practices of Building Machine Learning Systems
summary
Argues that 80%‑style failure rates persist because tooling alone does not solve operational complexity, and describes building a recommender platform for Storytel that adds orchestration, monitoring, GPU-backed training and experiment infrastructure. Finds concrete failure modes: data drift causes catastrophic accuracy loss in days, missing model versioning breaks rollbacks, long feedback loops (hours to weeks) foil quick fixes, and A/B tests need weeks to assess; platform reduces incidents but does not eliminate human and cultural work.
AI-written summary · about
chapters (6)
- 0:00Start
- 7:08What is a machine learning system?
- 12:41the problems start
- 28:14Resist Big Tech FOMO
- 31:52The hardest problem is not the tech
- 40:11Feedback loops vary from seconds to weeks
related talks
27mKeys to Building Machine Learning Systems
08mConsiderations for producing the best AI and Machine Learning models
18mReliable Maintenance of Machine Learning Models
29mDemystifying Machine Learning in Production: Reasoning about a Large-Scale ML Platform
39mMachine Learning Engineering
32mReliable Machine Learning
30mChallenges of machine learning: The new ML lifecycle
59mWhy Great Models Fail: Lessons From 9 Years of Deploying ML Models