Next week I’m due to begin a job as a Machine Learning Engineer at a company that works in the MLOps field. It’s a new field to me. I’ve read a good deal on it in recent weeks, and listened to a few dozen episodes of the MLOps.community podcast, but I still very much consider myself a beginner in the space. To that end, I thought it worth clarifying my understanding of what MLOps is all about, the problem it is trying to solve, and where I see the opportunity there.
A top-down explanation is probably the best way to think of what we’re doing when we talk about ‘doing MLOps’: we’re doing all the things which make it possible to train, deploy and use machine learning models in the real world or ‘in production’. It isn’t just a series of tools, but also a series of best practices and a community that is constantly learning and iterating to improve.
The kinds of things that you can do with machine learning models are incredibly diverse, so it stands to reason that the people who operationalise all these models have quite varied opinions and approaches to how best to do this. Even the deployment scenarios are pretty different and involve different technology stacks. There is an idea of a ‘full stack machine learning engineer’, which apparently means someone who just knows everything across the board; I hope to be able to delve into some of these areas and the key technologies represented in each space in due course on this blog.