One ofthe biggest challenges many companies face today is the gap between what candidates learn in university and what roles actually require in the real world. At the same time, recent graduates feel underprepared, as much of education focuses heavily on theory, with limited opportunities to apply knowledge in practical settings. As a student in the Master in AI for Business at EADA Business School, this gap is something we actively try to bridge through hands-on learning experiences.
To help address this challenge, the Program Director of EADA’s Master in AI for Business, introduced a hands-on challenge: a DATATHON. Students were grouped into teams and tasked with competing to develop the best-performing AI solution.
Preparing for the Datathon
Going into the challenge, we felt a mix of excitement and uncertainty. Building a model from scratch was something none of us had prior experience with, and the only thing we knew for sure was that we would be working on a medical imaging model.
Looking back, the preparation for this challenge actually started much earlier. In the first trimester, we began with coding classes that helped us get comfortable working on larger, more complex projects. This was followed by Machine Learning, Artificial Intelligence, Deep Learning and Large Language Models (LLM) courses, which built the foundation needed to start thinking about real-world applications. Three days before the competition, we attended an intensive three-hour session focused on how to approach the modeling process and understand how medical imaging models work. The session was led by a professional working in the field, which made a significant difference.
Rather than feeling like a traditional lecture, the class became an engaging discussion. We exchanged ideas, asked questions, and gained practical insights. He guided us through the modeling process step by step, clarified our doubts, and ensured we had the fundamental knowledge needed to begin building a model.
Building an AI Solution Under Pressure
The rules were clear from the start: we had four days to build a model, develop an application, and prepare a business plan. On the fifth day, we would present everything to a panel of judges.
Our main objective was to build an AI model capable of detecting cardiomegaly. In simple terms, cardiomegaly refers to an enlarged heart. While it is not a disease in itself, it can lead to serious complications, such as stroke, if left undetected. None of us had medical experience on the matter, so the first course of action for each group was to research the condition in depth—how it is diagnosed and how AI could be used to detect it more quickly and effectively.
We were provided with a dataset of frontal chest X-rays, divided into training and test sets, with each patient associated with a single image. Each day, we submitted a CSV file containing the image filename, the predicted probability of cardiomegaly, and a binary classification (0 or 1). The challenge was that the test dataset was not labeled, so we had no immediate feedback. Instead, performance was revealed through a daily leaderboard ranking teams based on metrics such as AUC, sensitivity, and specificity.
For the modeling, we were allowed to use pre-trained models, as long as they had not been trained on medical imaging data. This meant each team had to research and implement the most suitable architecture within a very limited timeframe.
At the same time, we developed a graphical interface capable of processing X-ray images and returning predictions. Since the application was intended for medical use, it needed to be intuitive, fast, and reliable.
In parallel, we built a business plan to support our solution. This included defining the target market, identifying the need for such a tool, and designing a go-to-market strategy for clinical deployment. This part was particularly challenging, as it required us to think beyond the technical solution and consider infrastructure, market demand, and financial viability.
Presenting Our AI and Business Solution
By the final day, the stakes were high. Every team aimed to deliver not only a strong model, but also a convincing application and a solid business case.
We began our presentation with the business plan, explaining our chosen market, the problem we aimed to solve, and the associated costs and infrastructure requirements. We also presented our financial projections, including a path to profitability and a projected break-even point within three years.
We then moved on to the technical component, explaining the logic behind our model and the decisions we made throughout the development process. Finally, we demonstrated our application and how it could be used in a real clinical setting.
Afterward, the judges asked questions to assess both our technical understanding and our ability to apply key deep learning concepts.
Key Takeaways from the EADA AI Datathon
Overall, the week was intense, challenging, and incredibly rewarding. Although we were competing in teams, there was a strong sense of collaboration throughout the process. Teams shared ideas, exchanged tips, and supported each other along the way. While this experience does not fully close the gap between academic theory and the job market, it really does help to create momentum and motivation to keep exploring the matter in practical ways. It also makes us feel better prepared for the roles we aspire to pursue in the future.
On a personal note, I am especially grateful for my team. What started as a group formed on preparation day quickly turned into a strong and collaborative unit. And in the end, it paid off—we won the datathon.
So, to the Strawberry team: you rock.
Author
Nereid King
Participant - International Master in AI for Business 2025-2026