Ruben Ahrens, MSc

In collaboration with the BI team at NOC*NSF, I worked on a project to analyze sports participation data in the Netherlands. Using Databricks, we processed large datasets, including CBS socioeconomic data to gain insights into sports participation trends for various neighborhoods and regions.

In a team of full stack developers, I collaborated on the "Axoniq Developer Agent". This web application enables developers with limited knowledge of the Axon Framework to create event-sourced applications using a GUI. My responsibility was mostly on prompt-engineering, and designing the agentic sytem, however, I also contributed to the backend and frontend development.

For the Dutch rowing federation (KNRB), I rejuvenated an outdated dashboard that visualized rowing competition data. The dashboard is now available accross the organization as a GUI web application.

My first engagement for Data Science Agency involved developing a proof of concept of a RAG agent for the Dutch Olympic Committee and Sports Federation (NOC*NSF). The agent allows athletes to ask about nutrition and get nutrition and training advice depending on their personal training data.

Earth observation helps monitor shipping emissions. This study uses machine learning to improve ship plume detection by incorporating SO$_2$ and HCHO alongside NO$_2$ from TROPOMI data. An XGBoost classifier trained on 80x80 km samples shows that adding SO$_2$ and HCHO enhances detection, especially at extreme NO_x proxy values. Individually, SO$_2$ and HCHO achieved ROC AUCs of 0.647 and 0.634, compared to 0.684 for NO$_2$, highlighting their potential despite room for improvement with more data.

In this course paper for the course "Sports Data Science", I combined my passion for bodybuilding and AI. I explored training data from 60 people, creating an algorithm that translates weightlifting performance across exercises.

Through fine-tuning on the CADEC dataset, consisting of medical reviews, the transformer network BioBERT, specialized in biomedical texts, demonstrates high performance in recognizing medical entities.

In this project, my peers and I used a genetic algorithm to optimize the container placement on a cargo ship that had a route passing multiple harbors. The placement was meant to minimize unloading time and the distance between the center of gravity of the ship and the load.

This is the first big project where I worked with spatiotemporal data. Me and my partner divided the map of Chicago, Illinois into grid cells, creating a sparse 50x55px image. Using an ensemble of ConvLSTM models, we were able to generate an accurate estimation of high-crime locations, accounting for high differences in data for different regions.

My bachelor thesis was my first big AI project. In this project, I used the deep learning model CLIP to classify paintings into portrait or landscape classes. I investigated the usefulness of interactive machine learning (where an algorithm is trained during the data annotation process). I created a GUI application to annotate the images.

For a course at the University in Amsterdam, I collaborated to create an educational VR game. In "Free the Sea" a player is tasked with collecting plastic waste in the sea. The purpose of the game is to raise awareness about recycling among children.