MLOps with Databricks

Chapter 1. MLOps Principles and Components

Machine learning operations (MLOps) is a set of practices that help to bring ML models to production in a scalable and efficient way. It can be seen as an extension of the development and operations (DevOps) philosophy that started around the end of 2000 as a movement against the established, traditional software development model, in which code developers and developers who deploy and support the code are separated functionally and organizationally. Siloed teams—with their own planning and key performance indicators (KPIs), inability to communicate code to the operations team, and misaligned development and production environments—contributed to long release cycles and created endless frustration.

MLOps addresses the same core challenge as DevOps did: building systems that can be reliably deployed, operated, and improved over time in production. The difference is that AI systems introduce additional parties—such as data scientists, ML engineers, and other specialized roles—as well as new types of challenges related to evaluation, testing, and reproducibility.

Unlike traditional software systems, ML applications are built not only from code but also from data. Their behavior is inherently probabilistic, which makes it significantly harder to define expected outcomes, write meaningful tests, or validate system performance before deployment.

These characteristics make the development lifecycle of ML systems fundamentally different from ...