The Culture, Architecture and Technology of MLOps
summary
Argues that putting models into production is mostly culture and process not magic, equating MLOps to DevOps plus machine learning and insisting teams share responsibility; lays out a reference architecture with model registry, feature engineering, online feedback loops and scheduled retraining (batch weekly/monthly to continuous online), and concrete plumbing: feature store (Feast, Hopsworks, cloud providers), workflow orchestration (Airflow, Kubeflow, MLflow), model-as-microservice vs in-stream serving (PMML/ONNX), and warns there is no silver bullet—start small and scale.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:07What is MLOps
- 6:04MLOps culture
- 9:42Core architecture for machine learning in production
- 16:51Monitoring and CI/CD
- 20:29MLOps tools
- 27:11Generative AI serving
- 40:42Key takeaways and further resources
- 42:29Q&A
related talks
41mTHE STATE OF MLOPS - MACHINE LEARNING IN PRODUCTION AT ENTERPRISE SCALE
30mMLOps for Continuous Learning
13mMLOps at Scale
53mExperience with an end-to-end MLOps pipeline: Practical insights and 3D architectural perspectives
01mMLOps: Applying DevOps to ML
26mAre you ready for MLOps? 🫵
31mMLOps In Practice – How To Run Your Machine Learning Models In Production At Enterprise Scale
36mIs the MLOps tooling the same as the DevOps tooling?