MLOps (Machine Learning Operations) is a multidisciplinary practice that bridges the gap between ML development and operations. It aims to streamline the process of designing, developing, deploying and managing ML solutions in the real world. MLOps involves a myriad of techniques such as continuous integration, continuous delivery, and automated ML. These practices are derived from DevOps, a set of practices aimed at shortening the system development lifecycle, and are now tailored to meet the specific demands of Machine Learning.
MLOps is a fairly new term in the AI field. It emerged around 2015, in pace with the rise of digital technologies, and started gaining popularity in the latter half of 2010s. The term was born out of necessity, to address the challenges that arose when differentiating between the traditional software engineering cycle and the unique requirements of a Machine Learning pipeline - these involve data versioning, model evaluation, model versioning and model monitoring among other things.
While no single individual or group is credited with the creation or popularization of MLOps, it is an evolution of the DevOps concept with specific adaptations to accommodate the unique characteristics of Machine Learning workflows. Tech giants like Google and Microsoft have been instrumental in pushing the MLOps practices forward, by building and promoting MLOps tools and best practices.