It could be the title of a children’s book: Jeffrey Luppes and the model of a Machine Learning Engineer.
A world of algorithms, anomalies, and attributes awaits. But this is no fairy tale nor sci-fi story; it’s reality – sometimes augmented. To grasp what it means to be a Machine Learning Engineer, we went on a data date with Jeffrey.
(Data) Set the Scene
Like any other company with many colleagues, we try to make sense of our responsibilities and dependencies with an organizational structure. In our case: with clusters, areas, squads, and more. For Jeffrey, that means that he is part of the Data Team within Online Services within DPG Media. The Online Services cluster services over ten brands – you probably know quite a few of them: Tweakers, Independer, Nationale Vacaturebank and Autotrack. And the data team is involved in everything related to data for and of these brands.
With twelve individual brands, you can imagine that there are many technical challenges to deal with. Luckily, the team offers a central platform for the brands to use. The team always tries to find common denominators and synergies between Online Services brands and other DPG Media brands. The team consists of Data Engineers, Data Scientists, a Business Intelligence Analyst, and Machine Learning Engineers like Jeffrey. He has a varied job: he creates models and deploys them, improves the monitoring for existing machine learning models, talks to stakeholders about their needs and wishes, and much more
“I’m the link between data science and engineering. My main responsibility is to make sure we produce machine learning models in a way that is not only correct but also maintainable. Machine learning models are basically predictive functions that help our users. For instance, we might build models to make it easier to find the perfect job based on your search queries. Or let’s say we want to give all articles on a website a tag; we then create a model that labels all articles correctly. To illustrate: if we have a classification for ‘consoles’, all articles on consoles should receive that tag. In general, when one of our Data Scientists or I create an awesome artificial intelligence concept, I make sure that we can actually use it, that we can push it live.”
Engineering is Pioneering
It’s pretty common within DPG Media that Data Scientists and Machine Learning Engineers have a strong background in programming. So does Jeffrey. “After my bachelor’s in software engineering, I worked as a developer. Around that time, 2015, I believe, data science was becoming ‘a thing’ that universities were picking up on. I worked with huge data sets as a software engineer, and I was very intrigued by the work done by the people who created those data sets. At the time, these were statisticians, econometricians, and mathematicians – nowadays, we’d just call them Data Scientists. I wanted to get in on it, so I went back to the university for a master’s in data science.”
It took some time and dedication to get used to the student lifestyle again – especially when it was time for Jeffrey’s graduation project. Jeffrey smiles: “I was doing the exact same thing as an intern as a regular employee – only now for a meager internship allowance.” Luckily, the company hired him as a full-time employee straight after his internship. After a few years, though, Jeffrey decided to take on the Machine Learning Engineer position at DPG Media.
What brought him to us? “I love working with machine learning in production and was looking to get more experience with tech and platforms like AirFlow and AWS – and that’s exactly what DPG Media does. I also like that my job is somewhere between science and engineering, which allows for plenty of pioneering and figuring things out.” And there is a lot to figure out. For example, he’s now racking his brains on how to calculate CO2 emissions of the computing power the models use, a request by a Data Scientist. So like Pippi Longstocking famously said: “I have never tried that before, so I think I should be able to do that.”
Jeffrey started his career at DPG Media amid the pandemic. At first, it felt a bit ill at ease getting to know his teammates and the brands. “My whole team is used to working remotely, so for the work aspect, it doesn’t matter. But social bonding is definitely a challenge. It took me six months to meet my entire team in person. The daily stand-ups aren’t really sufficient for social talk, so we also plan a coffee date now and then.”
If you have development experience and you can work with Python, you’re well on your way to becoming a Machine Learning Engineer.
Tech Stack with Lots of Freedom
The team is brand new and still defining itself. They have to advertise themselves a bit: explain what they can do, take away the brands’ doubts and insecurities, and demonstrate what technologies they use. “A lot of Amazon Web Services, MLFlow, TensorFlow, Terraform, and soon we’ll switch to Grafana for dashboards. And Python, we use Python for basically everything.”
There is plenty to be proud of already. For example, the data team and Tweakers recently worked on a project to provide all articles with the correct tags. This is extremely important for advertising purposes: it creates more advertising space on the one hand (as all articles now have tags), and it ensures that ads are more relevant to the article. “I realize it doesn’t sound very flashy, but it’s a very practical application of machine learning. We start with the low-hanging fruit and upscale and expand from there.”
And as Jeffrey said, he and his team love pioneering, so they made their own framework for MLOps. MLOps is a set of practices that combines machine learning, DevOps, and data engineering. It helps them quickly train, deploy, and track their models, and because of the framework, the infrastructure on AWS is ready-to-go. Jeffrey: “You can obviously also buy MLOps platform solutions, but we wanted to have a flexible solution because we have so many brands and environments to integrate with. At the same time, we can’t build and manage everything ourselves, so we had to strategically choose managed services in some spots.”
Jump-Start a Career in Machine Learning
Any advice for people who want to get into machine learning? Do they have to do a Master’s, like Jeffrey did? “Maybe not anymore. If you have development experience and you can work with Python, you’re well on your way to becoming a Machine Learning Engineer. A lot of this stuff simply isn’t taught at universities.
There is a lot of information and inspiration available online on Reddit and Medium – also, definitely check out the MLOps Community. From there, it’s a matter of gaining experience at a company like DPG Media.”