Implementing different MLOps maturity levels
Most new ML teams and organizations go through a phased MLOps journey as they build and refine their MLOps strategy. They usually start with a fully manual step-by-step process where data science/data engineering teams take an extremely manual, ad hoc approach to building and deploying models. Once a few models have been deployed and stabilized in production, it slowly becomes apparent that this manual process is not very scalable and that the team needs to put some processes and automation in place.
At this point, as issues arise in production, it also becomes apparent that this ad hoc approach is not easily auditable or reproducible. As the usage of the ML solution grows, it graduates from being just an experiment to something the organization becomes increasingly dependent on. Compliance teams and leadership also start making requests to make the model deployment process more well organized and auditable to ensure compliance with the...
