In part two of this series, Demetrios Brinkmann, MLOps Community’s Head of Community and the MC of Tecton’s apply(conf) event, shares his perspectives on the greatest challenges currently in MLOps, the role of community in solving them, and more.
Note: The interview was edited and condensed for clarity.
Community isn’t the first word that comes to mind when you hear the word “MLOps”, but it might be the most important. As the field of machine learning operations (MLOps) continues to evolve and new challenges with bringing ML models into production arise, the best way for data scientists, ML engineers, and others to learn and address the challenges successfully is through communicating with each other.
Since the start of the pandemic, MLOps Community has helped bring them together through different forums, including podcasts, Slack, meetups and more.
A community of 10,000 today, MLOps Community is sponsored by some of the leading names in data and ML, including DataRobot, Redis, and Tecton.
RTInsights recently spoke with MLOps Community Head of Community Demetrios Brinkmann and MC of Tecton’s recent apply(conf), about the greatest challenges currently in MLOps, the role of community in solving them, apply(conf), and more.
Q1: You’re the Head of Community at MLOps Community. Can you talk about it and your journey to the MLOps Community?
My journey has been kind of random. And there’s been a lot of serendipity. After I graduated college– and I graduated in Spanish literature and Portuguese – I moved to Spain to teach English and enjoy the Spanish culture and just be in Europe.
And then, fast forward a few years, my daughter was born and I decided to do something that wasn’t teaching English because teaching English was basically a job for nine months of the year, and then, after those nine months, during the summer, everyone went to the beach, and they didn’t want to learn any English at the beach.
I had a friend who had gone into sales. And so I went into sales [and] I started out at a password management company.
And then after that, I ended up working at this company that was called dotscience. And they were doing MLOps. They were building an MLOps platform. In 2020, right before the pandemic was official in the West, people stopped wanting to have any kind of sales conversations… And so that led our CEO at the time to say, “Hey, you know what, maybe we could do something around community.”
He gave the first talk. [From there] we started to organize different things. I started doing weekly meetups, and I just started running them like podcasts… And then I started to learn what pain points people are having, and then it resonated with more people. And the community started to grow and it started to be something that was just a beast in its own right. And here we are.
Q2: Right. What are some of the pain points of the community?
I think the pain points right now are like, this is a very new field. And there are new people coming into it from very diverse backgrounds. So you have people coming into it from the data science background, or even like, “Oh, I was a physicist, and then I became a data scientist. And now I’m tasked with putting my models into production. And so I have to figure out this engineering side,” or you have someone who is a data engineer.
Related: Git-based CI/CD for Machine Learning and MLOps
And now they’re starting to learn more about why data engineering is different when you’re dealing with machine learning. Or you have someone who is coming from DevOps, and they have these strong DevOps principles, but they need to understand what’s different when you add data into the mix and how things can be unreliable.
When data is there, how you can version data, how you can make sure to keep like, PII like security is a huge one.
So this is from somebody who’s coming from DevOps, it’s very familiar to them. But when you are dealing with data and you’re dealing with machine learning, there’s just more complexities that are added to the mix. And so I think the range of backgrounds that you can have coming into the MLOps space is quite difficult and then it doesn’t help that this is such a new category. And it’s such a new field. And there’s not a clear definition or design patterns. It feels like nothing’s clear right now.
[For example,] you don’t have a clear stack or way of doing things or even use cases. Like, there’s things that are starting to emerge, maybe like recommender systems. And you can say, “Wow, okay, I’m starting to see so many papers being put out by the Facebooks and the Ubers, and the Airbnbs and those top tier companies that are talking about recommender systems,” that now you can say, all right, there is kind of this pattern that’s happening, or deployment patterns. And so there’s patterns that are starting to emerge, but still, it feels like everything is shrouded in mystery. Still.
Q3: It makes the community all the more important. You mentioned DevOps, and I saw your [blog] post from a few years ago about the difference between DevOps and MLOps, has that changed at all in the past two years?
The biggest thing that I think has changed is that there’s been an incredible book that’s been released from just some amazing people. It’s called Reliable Machine Learning. And there’s a ton of authors that I couldn’t say all of them.
The other thing that’s changed is probably that there is more understanding of it, just more at bats, from people who are in the DevOps field, they have been able to play around more or touch the machine learning and MLOps space more.
So there’s starting to be this understanding a little bit more.
As far as the biggest change, the Reliable Machine Learning book is just incredible. Two of the writers of that book are DevOps through and through and they went into MLOps, and they’re helping forge the new path.
Q4: So going back to what you mentioned, going back to what we were discussing about community before, and in all of these pain points. Can you talk a little about the Tecton conference? And what is the value for the MLOps community?
In my mind, this type of conference, you don’t really see in this space. I feel like it’s pretty clear. Because of the amount of registrants that it gets, because of the amount of success that it has, because of the amount of people that have reached out to me and said, “Wow, that was so valuable.” Because of the Tecton team, [who] work hard at making sure everyone who is part of the conference is someone that has something to teach you.
And also when it comes to that maturity level, or just that experience level. They’ve had experience. And so they’re able to show you things [from] their learnings. it’s not only that they’re farther ahead, it’s that in their specific use case, they’re opening up. They’re being very transparent about what they did and how they did it. And [in] a very technical manner.
And that helps inspire the rest of us when we attack problems and challenges in our day to day. And so yeah, the main thing that I think is key and is crucial here is that there’s incredibly brilliant minds that are coming.
And they’re being very transparent and very technical, they’re getting down to a deep level for a specific set of problems. And they’re showing others how they did it. So that it inspires us or the rest of us when we have to tackle similar problems, or maybe not similar problems just adjacent problems.
Q5: Right, I hear a lot about so much content out there that is supposed to be for engineers and it’s not as technical as they would like. So that’s great to hear what you said about Tecton’s conference, that it digs deeper into things.
I think that’s one of the highlights. That’s why people enjoy it so much, is because of that technical depth that is expected of the speakers and the hard work that’s put in finding these speakers.
Q6: What advice do you have for someone starting out in MLOps engineering?
Yeah, in ML, I would. Of course, it’s like a hammer looking for a nail. But I do find the power of community to be very helpful, especially in these new and very emerging fields. Because you get to have access and speak to and make relationships with people who are potentially doing things that you have to do.
And so I’m always amazed [at] the people that we have there, and how generous they are with their time and their knowledge. So I recommend joining a community. There’s a ton of resources right now, but there’s also a lot of noise. So trying to cut through the noise is a pretty hard task to do. That’s another reason why I think community is really nice because you’ll see what bubbles up, what’s important.
Read the first article in the series here.
