Governed MLOps on Google Cloud.

Most machine learning models never reach production, and many that do are switched off a year later. This book is the path that gets them there and keeps them there, with every command run on a live Google Cloud project.

• Data contracts, point-in-time features and reproducible runs
• Synthetic data that is useful and provably safe to share
• Pipelines on Agent Platform, Kubeflow on GKE, Ray and BigQuery ML
• Promotion and serving: canaries, A/B tests and rollback
• Drift and monitoring, with retraining a person approves
• CI/CD in GitHub, with every approval on record
• Lineage and evidence your auditors can read
• An endpoint behind a web app, with Apigee and Cloud Armor advice
• Coding agents that build use cases without seeing a row of data
• Running it all on GKE: Kubeflow, KServe and KubeRay

Customer segments, a demand forecast, lapse risk, a checkout offer and synthetic customers, on the Online Retail II dataset. Every chapter's finished code is in the example repository, one folder per chapter.

The PDF is free. Copy it and share it with your team or anyone else, unchanged and free of charge, with credit to the author, under the Creative Commons BY-NC-ND 4.0 licence.

Written by Rakesh Gupta, the founder of Pipewright, after more than twenty years building software for airlines, banks, automotive companies and governments.