Projects
2024
- In collaboration with a well-known german history podcast, I developed a chatbot that was capable of leveraging the knowledge from nearly 500 episodes. To this end, a machine learning pipeline was deployed to the cloud to automatically transcribe podcast episodes and organize the contents in a vector database. Using a RAG agent and a large language model (LLM) the system provided precise answers to detailed questions and could even pinpoint, who was speaking, when he was speaking and in which episode. Moreover, a dashboard featured useful statistics such as speaking time analysis and word clouds. read more
- This was a project for a big german mobile provide with over 6 mio. customers. The goal was to better understand customer needs and satisfaction based on what they were reporting, when they were calling the companies' service hotline. To this end, transformers and LLMs, speech-to-text models and natural language processing techniques were used. Finally an end-to-end ML-Pipeline was deployed, delivering new and valuable insights into their customer's need on a daily basis. read more
2023
- Development of a product for contextual - and 3rd-party cookie independent - targeting on websites. My task was to build data preparation, processing and training pipelines, as well as deploy inference endpoints for their machine learning models in the cloud. Additionally, a web application was developed to simplify the creation of new contextual customer segments by providing interactive tools to foresee effects of different configurations, ultimately leading to a better understanding of the inner workings of the underlying machine learning models. read more
2022
- The goal of my master's thesis was to investigate, to what extent machine learning models were capable of automating tactical analysis in football to benefit coaches and decision makers. Spatiotemporal data (players' and ball's xyz-positions) from 15 professional first-division Bundesliga matches were obtained and manually annotated regarding the respective tactical maneuvers that were employed by the teams. Using this feature-engineered dataset and supervised learning techniques, machine learning models were trained to accurately predict these tactical elements. read more
2021
- Using computer vision and a convolutional neural network (CNN) a model was trained to play the game Subway Surfers. It learned from my own playing, through which I have created and augmented a dataset from which it could induce the best moves to make based on what was happening in the game in real-time. read more
- For this project I have programmed the game Icy Tower and used reinforcement learning and a genetic algorithm to iteratively train neural networks to master it. After many generations of neuro-evolution the algorithm produced genomes that played the game better than I could. read more
