Machine learning engineer: the job of a problem-solver

As a teenager, Ahmed Ahres dreamed of making virtual reality gaming software. Today, he is developing facial recognition technology to prevent financial fraud at Revolut. He gave us a behind-the-scenes tour of his career as a machine learning engineer, at the Canary Wharf-based start-up.

How did you develop an interest for machine learning engineering?

“When I was younger I was big into gaming and I wanted to develop gaming software and particularly virtual reality (VR). I moved to Eindhoven in the Netherlands from Tunis in Tunisia when I was 16 and studied my Bachelor’s in Computer Science. I was really excited by the application of VR with computer vision and its potential. So I continued with a Masters in Computer Science concentrating on VR and computer vision at Lausanne in Switzerland.”

You started at Revolut as a data scientist intern in financial crime. How did you get the internship?

“Revolut didn’t have an intern programme but after some networking, I was given the chance to sit a round of interviews with the prospect of an internship at the end. After 5 interviews, I finally got a spot.”

As a data scientist, I compiled data to help build AI models. The models are used to detect and prevent fraudsters who are trying to impersonate customers.

What did you do as a data scientist intern?

“As a data scientist, I compiled data to help build AI models. I would get the data “model-ready” in order to train the AI. The models are used to detect and prevent fraudsters who are trying to impersonate customers.”

Machine learning engineers are responsible for designing…models to solve real-world problems.

You then scored a full-time position at Revolut as a machine learning engineer. What does your job title mean?

“Machine learning engineers are responsible for designing, creating, evaluating and producing models to solve real-world problems. Nowadays, companies have in hand a large amount of data with which engineers build models. For instance, by having access to sales numbers of a certain video game over the years, we are now able to predict what would be the number of sales in the upcoming year. A machine learning engineer would then be responsible for designing, creating and evaluating such a prediction model.”

How is their job different from that of a data scientist or data analyst?

“Machine learning models require data to be able to train and be evaluated. The machine learning engineer approaches the problem assuming that the data is available. One of the tasks of a data analyst or data scientist is to provide such data.”

I train the AI to know if the selfie of one of our users matches the passport photo of the same user.

What are your day-to-day responsibilities and duties as a machine learning engineer at Revolut?

“I work on developing algorithms and training our machine learning models to recognise and match faces of our customers and to block those who aren’t. This involves a lot of data modelling and a lot of AI training. For example, I train the AI to know if the selfie of one of our users matches the passport photo of the same user. Or more specifically, when it doesn’t and when that might be a potential fraudster trying to access a customer account.”

“At the same time, my team is small. So I take up tasks that are outside my role. For example, we want to do some analytics. I’m not a big fan of analysis but the team needs it, so I have to do it.”

Your team—Computer Vision team—is quite new at Revolut. What are the other departments you work with?

“We work a lot with the Know Your Customer team (KYC) and collaborate with departments such as Financial Crime and Compliance. We are building and testing facial recognition so this is a key area.”

To me, the most amazing thing about being a machine learning engineer is the problem solving process. This for me is more important than just the technical side.

What’s your favorite part of the job? And what’s your least favorite?

“To me, the most amazing thing about being a machine learning engineer is the problem solving process. When given an important problem to solve, you have to ask yourself important questions such as, “How will I design the model and why?”, “What is the smartest way to validate the model?”, “Which data do I need and where will I get it?” etc. This for me is more important than just the technical side.”

“Particularly at Revolut, I was given a lot of trust and responsibility. This motivated me a lot. When I joined this company, I was still 20. But as soon as I joined, I was given a big project. Within three weeks I was interviewing others for the team and I was influencing the code base straight away.”

Machine learning is more than just a technology but a mindset when it comes to solving problems. Building your own projects proactively, taking online classes will teach you that.

What advice would you give to someone looking to become a machine learning engineer?

“There are different ways in which someone can become a machine learning engineer, apart from studying computer science. I would say to be proactive and do online machine learning competitions such as Kaggle competitions. Kaggle is an online platform with real-world prizes and real-world challenges presented by companies where people have access to the data and are asked to build machine learning solutions.”

“You could also take online classes such as the famous “Machine Learning” class on Coursera by Andrew Ng and then create personal machine learning projects using open-source data. By publishing those projects online (for example on a personal website), one is able to show his/her skills and how proactive he/she is. Machine learning is more than just a technology but a mindset when it comes to solving problems. Building your own projects proactively, taking online classes will teach you that. Of course, one should also learn technical skills such as TensorFlow & Python.”

Photos: Betty Zapata for Welcome To The Jungle

Follow Welcome to the Jungle on Facebook and sign up for our newsletter to receive our articles every week.