MLOps: Definition and Examples

MLOps is a discipline at the intersection of machine learning, software engineering, and IT operations (DevOps). Its main goal is to bridge the gap between the experimental phase of an AI model — developed by data scientists — and its deployment in a real-world environment, where it must operate reliably, performantly, and maintainably. Concretely, MLOps covers the entire lifecycle of a model: training data management, versioning of models and datasets, automation of training pipelines (CI/CD for ML), continuous deployment, and production performance monitoring. Without MLOps, companies face the well-known problem of "the model that works in a notebook but never in production." One of the specificities of MLOps compared to classic DevOps is managing "data drift" and "model drift": real-world data evolves over time, which can degrade a model's performance without any code change. MLOps therefore integrates continuous monitoring systems that detect these drifts and automatically trigger retraining if necessary. In the context of generative AI and prompt engineering, MLOps now extends to "LLMOps": managing prompt pipelines, versioning templates, systematically evaluating generated responses, and optimizing inference costs. These practices are essential for any organization that wants to integrate AI in an industrial and responsible manner.