Platforms for valuable AI Products- Iteration, iteration, iteration
summary
Argues that rapid, repeated experiments are the path to value, with concrete tradeoffs and tooling choices: use Google MLOps stage two for a reference architecture, prefer incremental slices and shadow deployments, build a continuous training pipeline, and serve models as docker + FastAPI + scikitlearn predict schema. Cites outcomes: handling billions of daily events, moving deploys from a week to an hour, automating daily retrain, and recommendation models generating 80 million pounds per year as evidence of what actually works.
AI-written summary · about
chapters (24)
- 0:00Introduction and objective
- 0:56The importance of iteration
- 2:17Speaker background
- 2:56Lessons from past projects
- 4:02Iterative design cycles
- 5:05Architecting for success
- 7:12Avoiding big bang failures
- 8:17Probabilistic systems
- 9:35Experiment value curve
- 10:05Case study on failure
- 11:31Building experimental systems
- 13:04Data infrastructure
- 14:10Automation and standards
- 15:30Continuous training pipelines
- 16:21Evaluation and observability
- 17:10Retailer success story
- 17:42Production and dev boundaries
- 18:58Shadow deployments
- 19:30Fallbacks and rollbacks
- 20:18Feedback and culture
- 21:41Cross-functional teams
- 22:30Balancing maintenance and growth
- 24:19Leadership and culture
- 25:15Conclusion
related talks
33mAI with a DevOps mindset - experimentation, sharing ..
28mLeading machine learning research journeys with happy (into-production) endings
11mSuccessful AI adoption
26mMLOps for the rest of us- A poor man's guide to putting models in production
44mAI workflows and teaming - how do teams deliver AI products?
13mMLOps: From AI Research to Production-Ready Systems
41mFrom Prototype to Production Building a Scalable MLOps Platform from a Reference Arch
26mCompleting the machine learning loop