MLOps That Survives Production · Kai London
Registries, Feature Stores, Monitoring and Retraining
Your model passed validation. Will it survive the clock change, the supplier feed and the 3 a.m. page?
No downloads, no subscriptions, nothing extra to buy: every numbered item in this book is printed in full and completed for a worked organisation, so you can adapt it to your own organisation straight from the page.
Premium Practitioner Edition. A complete working reference, not a quick read: practitioner chapters, a field toolkit of worked case studies and labs, and a reference section with practice questions, checklists, an executive summary and an operating model you can adopt.
Your model passed validation. Will it survive the clock change, the supplier feed and the 3 a.m. page?
Most machine learning models do not fail in the notebook. They fail months later, in production, while every infrastructure dashboard stays green: skewed features, a supplier that changes units without changing the schema, a canary that missed the failing region, an owner who left. This book gives you a tested method for keeping models reliable, reproducible and governed after they leave the notebook.
Twelve chapters follow the life of a model, from the Production Gap Audit and model registries with promotion gates, through feature stores, reproducible training pipelines and data contracts, batch, online, streaming and edge deployment, and monitoring for drift, data quality and decay, to incident response and rollback, retraining with champion-challenger evaluation, regulatory evidence, platform economics and the operating model. Fifty numbered production case files, PCF-01 to PCF-50, are printed in full at one fictional retail, finance and logistics group with 43 models in production. Generative and retrieval services are covered too: prompt and index versioning, groundedness and freshness monitoring, and evaluation-set challengers.
Inside you will find:
• Fifty production case files, each ending in a completed artefact you can adopt
• Registry schema, promotion gates and alias rollback that make every model traceable
• Point-in-time parity tests, skew budgets and data contracts that stop silent failure
• Drift, data-quality and decay alerts, incident runbooks and champion-challenger tests
• Case studies, ten labs, thirteen working papers, a crosswalk and a 90-day plan
Every case file ends in a completed artefact: a model registry schema, a feature view design template, a data contract, a monitoring dashboard with Prometheus-style alert rules, a retraining decision record or an incident runbook with real commands and a completed run record. Chapter 11 indexes all fifty by failure pattern and symptom.
Part II is a field toolkit: two worked case studies, ten hands-on labs from registry gate to rehearsed rollback, thirteen completed and annotated working papers, a five-level maturity model across ten dimensions with a thirty-metric pack (the instrument for independent assessments), and a week-by-week 90-day plan. Part III adds a standards and regulatory crosswalk, six tabletop exercises, a sixty-question practice bank, thirty failure lessons with recovery steps and a tooling evaluation guide.
Who it is for: ML platform engineers, MLOps leads, heads of data science and the SRE teams who support models in production, together with the model risk, audit and data protection professionals who must evidence how those models are governed.
Part of the Professional Validation Library by Professor Kai London: practical books that turn standards, technology and regulation into work you can evidence. The print editions include an author's note, a ten-point readiness self-check that routes you to the right chapters, twelve framework figures, key takeaways and a dated "what changes next" section for every chapter, discussion questions on the case studies, consolidated checklists, a glossary of about 130 terms, self-test answers, a RACI matrix, a quick eight-dimension quarterly scorecard mapped to the ten-dimension maturity model, an afterword and a full index.
Professor Kai London