Senior ResearcherContact Markus
The development of trustworthy AI solutions - in most cases Machine Learning (ML) - requires appropriate engineering methods. MLOps is an emerging engineering paradigm that offers quality assurance in automated pipelines. RISE in Lund runs a testbed and demo environment for customers to "test before invest" in the booming MLOps market.
Anyone who tried migrating a promising AI proof-of-concept to production knows how hard it is. We embrace MLOps to overcome the hurdle through workflow orchestration - we make the complex AI pipeline plumbing look simple. Let us show why an MLOps pipeline is a prerequisite to quality assurance in the AI era.
MLOps is an engineering paradigm supported by tools and processes. Building on DevOps, MLOps is customized for the experimental nature of data science - adding experiment tracking and model management as first-class citizens. Successful application of MLOps increases the quality, simplifies the management process, and automates the deployment of Machine Learning (ML) models in large-scale production environments. Embracing MLOps makes it easier to align ML models with business needs, perform quality assurance, and demonstrate adherence to regulatory requirements. MLOps applies to the entire lifecycle – data gathering, ML model development, conventional software development, continuous integration/continuous delivery, orchestration, deployment, governance, and business metrics.
Workflow automation is the backbone of MLOps, i.e., the pipelining that makes AI real. Organizations embarking on the MLOps journey have two primary options.
There is not one single MLOps pipeline to rule them all. Each application needs its own customized pipeline - and it must co-evolve with the AI solution. At our local testbed, we help development organizations navigate the overwhelming options. We maintain a set of MLOps pipeline solutions that showcase the key variation points.