One of the many experts in the Medius team is Tadej Justin, PhD, Chief Data Scientist, who leads the ML and AI lab at Medius. We took this opportunity to sit down and debate with him about different tech topics, people, and how technology can help us solve certain challenges.
Between September 19-20 Tadej will lead a Databricks workshop held at the Faculty of Computer and Information Science, where he will share good practices and tools for data processing, data analysis, machine learning and AI. Find more about the event at the link.
Dr. Tadej Justin
Tadej, has data always been something you wanted to research?
I think that in life we tend to look for new challenges wherever we go and so you never know where the journey will take you. Currently, I’m working with data and machine learning at Medius, which offers me a lot of creativity as we're a boutique company offering boutique applications, and with that come boutique challenges. Each project is a story in itself, the challenges are always different, but we still transfer a lot of knowledge from one project to the next. I have, however, always been driven to research and discover the field of data science and machine learning and thus solve increasingly difficult challenges.
Before we continue, can you tell us more about yourself? How did you end up here at Medius?
Computer science caught my attention when I was attending Poljane High School. Latin was part of my curriculum and even back then I looked at the language differently, through the prism of how it could be processed with computers. After high school, I enrolled in the Faculty of Electrical Engineering, which was a great decision, as it gave me an even broader view of computer science. Electrical engineering gave me a different perspective on automation processes and helped me later when I focused on cybernetics and artificially intelligent systems. After finishing my studies, I got a position as a young researcher at the Faculty of Electrical Engineering and opened my own company together with some friends from university. The company still exists today, we deal mainly with language processing and web applications development. As a doctoral student, I expanded my interests in language and computer science by working in the laboratory for the artificial perception of systems and cybernetics, and I defended my PhD dissertation on the topic of Slovenian Emotional Hmm-based Speech Synthesis.
Your statements are full of playfulness, enthusiasm for understanding the language and the desire to use new technologies. Did you start using all that knowledge right after finishing your studies?
After finishing the young researcher program, I stayed on at the faculty for a while as a researcher, where I mainly dealt with speech and language, and recognition of emotions in speech. We also researched the possibilities of hiding the identity of the speaker, but it was niche research since we did all that in the Slovenian language. The Slovenian market is not big enough, corporations were not interested in our small nation at that time and did not show interest in expanding in this direction. Only years later did they adapt their bases and models for the Slovenian language, which is great. However, at the time the stakes in this field were very small, so I decided to upgrade my knowledge to a more general, more concrete and applicative part of machine learning and expand to other industries as well.
Because of your wide knowledge, you are torn between different roles. What does your job entail?
At Medius, I am responsible for a small team of experts who are working on machine learning and data science. We focus on exploring and learning new approaches. We are always moving between research and the realization of a client’s vision. When I came to Medius, we immediately started working on a methodology that would offer our insights and research in the field of machine learning to clients in a smart, standardized way. With additional programmers and a strong internal knowledge base we developed a platform for packaging our models and algorithms in an enterprise environment. Today, we are at a point where we can offer a consistent way of programming and implementing our insights and models for various projects. It is important to know that enterprise environments usually have their own infrastructure where public applications are rarely used, so trust in the existing infrastructure is on a high level. Our main challenge is to find a way to implement machine learning within their own environment, which in return increases the organization’s efficiency. We wouldn’t be able to do this without building a versatile and highly skilled team.
What technologies do you use?
Medius was traditionally a "Javanese" company where everything was based on the Java programming language. With the establishment of my team we had to expand this knowledge, because at that time Java did not offer the same support for machine learning as Python, which we know today as an open source program with a lot of support libraries for machine learning, post-processing and data visualization. The decision to start using Python also meant a big leap in the minds of the team which takes care of integration and upgrades. With this, we enabled a faster, more agile development process and wider possibilities for solving and researching even the most demanding challenges. Recently we’ve also been using the Databricks platform (Medius is also a Databricks Partner), intended primarily for our clients who use public cloud services like Google, Amazon and Azure. Databricks platform allows us to quickly process large amounts of data using Spark.
Let’s take a look at our local environment and the digitalization of Slovenia. How well is machine learning represented in Slovenian companies?
The Slovenian market is still in the phase of initiatives that encourage decision makers to educate themselves on using machine learning in practice. You can still feel the gap between what is feasible and what is expected. With a greater understanding of machine learning, we could have achieved more optimal results. We want to encourage decision makers and raise awareness which is necessary for these kinds of projects to come to life. We invest heavily in introducing machine learning and data science to the right people in innovative ways while, at the same time, presenting Medius as a reliable partner in the process of digitalization.
What about people? Your team, as you said, grows with the projects. How do you operate?
Each team member must understand the entire process from start to finish. From using data, learning models, and preparing algorithms, to using algorithms in production. Everything must be backed up with a lot of programming knowledge, knowledge of mathematical laws and operations, but also how to work with data on all levels. Our engineers and architects need strong programming skills, as we constantly encourage and educate the team in the direction where everyone has an insight into the entire process and technologies, which maintains a great environment. This is essential in my opinion, as we all realize how much effort is needed to successfully set up an application that performs in a production environment.
Many students would love to work here at Medius. What skills are you looking for?
In addition to basic programming knowledge, we are primarily looking for playfulness, the absence of stereotypes in the search for solutions and the willingness to learn about new technologies. Another thing we like to see in candidates are the soft skills that are needed for communication with the team and in programming in general. Above all, we certainly value ingenuity, which is key in our work. I invite everyone who recognizes themselves in this description to visit us, as there are still many challenges ahead of us.
The interview with Dr Tadej Justin was conducted by Bojan Miličić, Head of Business Development and Sales at Medius.