.svg)
.svg){kind=small}
“MLOps is so important when you start to get serious about the operationalization of your models. If you're just having one model in production, I mean go ahead, deploy it as you wish. But when things start to get serious with more than one model for more than one client, you need to retrain those things based on event patterns that come out of nowhere.”
Next up in our lineup is Tales Marra, Machine Learning Engineer at AssessFirst. He’s a true MLOps enthusiast, who hasn’t just incorporated it in his work, but writes about it on his social channels as well.
My name is Tales Marra and I'm a Machine Learning Engineer. I've been working in the data field for about four years now. I've done so many different things, I actually went from being a Data Scientist to more of a Machine Learning Engineer, then going back to being a Data Scientist. So I'm kind of in the middle of this flow in the data space that everyone knows so well. I also write a lot of posts about MLOps on LinkedIn because I think it’s fun.
I learned about MLOps about two to three years ago, when I worked in a company where we started to handle a lot of models in production. It became quite a challenge, especially when we migrated to deep learning models. So we had to start looking into cloud costs, especially to make our projects get some good ROI. That's when I started to see the importance of this field. I started investigating more and eventually I started investing my time in it. MLOps is so important when you start to get serious about the operationalization of your models. If you're just having one model in production, I mean go ahead, deploy it as you wish. But when things start to get serious with more than one model for more than one client, you need to retrain those things based on event patterns that come out of nowhere. Then you need to start thinking about taking this subject more seriously, especially when you have the pressure of the business side of things, which is always to basically demonstrate that the project generates value. I believe that generating value at scale is not something you can do without the proper infrastructure or management.
I'm an engineer so I like building things, and building things for AI is always fun. You can see your models running and the scale challenge is always fun. At least that’s what I like. It's interesting to see that you can do things at scale, but then you need to find the little optimization that will get you there to meet that challenge. So that's why I'm always on LinkedIn interacting with everyone. I'm always looking into if there are any new tools or things that we can do better and faster, so that we can ship things as best as we can.
Since I write a lot on LinkedIn and Medium, I'm very happy to talk to everyone and answer questions. But something I really recommend for everyone to take a look at are in general engineering blogs. There are many engineering blogs out there and many blogs outside of the scope of the big companies (that often tackle use cases at a scale that most other people or companies don’t encounter). So you probably should take a look at the engineering blogs of startups or different companies which are writing engineering blogs. I think that's very enriching for your profile as an engineer. So I really recommend the engineering blogs of Algolia, Linkedin and Spotify, and of course the Hopworks blog which I know you have a lot of good stuff on.
