It’s our ambition for Automated Machine Learning that building machine learning models should be an easy task where most of the technical heavy lifting is done by the platform. We therefore need to design a user interface which facilitates this. We spoke to Evangelia “Litsa” Mitsopoulou who joined Modulos in October 2019 to work on all aspects of the user interface.
Litsa, please tell us a little bit about yourself. What’s your background and what’s motivating you in your career?
Litsa: I am a front-end engineer. That means I build software for the Web. What’s motivating me in my career is the marriage between engineering and art. You directly see the impact of your goal directly on the browser. I really care about the user.
So what was it like to work with data scientists? Was there anything new about working with them compared to other types of engineers?
Litsa: At the beginning, I was thrilled by the potentiality of the product. It all sounded like a black box. It was a new language to me: so many new words and a completely different area. But working with data scientists has been quite smooth and interesting because it opens up a window to another world for me. And as we proceeded, I kept hearing about new datasets and I was fascinated about the wider applicability and usage of machine learning.
How did you approach building the AutoML user interface?
Litsa: I approached building AutoML as I would approach building any software: good quality and good performance. But on top of this, I take into consideration the industry-specific needs for a machine learning platform. That means we’ll have to deal with many real-time data updates on the interface. The second challenge is that we need to upload large datasets from different sources. That made the exploration and choice of the relevant tech stack really important.
What is your goal in building AutoML?
Litsa: I would like to be part of contributing to a solution which has a wider ethical impact globally. I was aware of more mainstream areas where machine learning was being used. So the more new datasets were brought on the table, I realized the greater applicability of our solution.