Projects

Research Projects #

Multimodal and Agentic AI for Learning Context Understanding #

For my dissertation, I work towards more scalable and generalizable learning analytics using multimodal and agentic AI. The first project built on embeddings from multimodal large language models to represent educational videos and predict population-level learner-video interactions.

Grammar teaching with LLMs #

My Master’s Thesis addressed the adaptation of LLMs for conversation practice with a focus on English grammar acquisition. The work was co-supervised by Mrinmaya Sachan and Peng Cui at ETH Zurich and Detmar Meurers at the University of Tübingen. You can find the resulting publication here. Preliminary work on how pretrained transformer-based embeddings represent grammar and text generation can be controlled in terms of grammar that I did at Yale, was featured at the BEA workshop at NAACL24 in Mexico City (see paper).

Knowledge Prerequisite Mining #

For my Master’s research seminar at the University of Tübingen, I investigated prerequisite structures of knowledge concepts and how to extract them from textbooks, Wikipedia and LLMs under the supervision of Álvaro Tejero-Cantero in the ML Science Colaboratory. Preliminary results have shown that early small language models such as T0 were not that good at this task. Given the recent advances of LLMs, this can be seen a bit of a museum piece of research.

Eye-Tracking for Educational Research #

As a research assistant under Sascha Schroeder, I improved the preprocessing of eye-tracking data from psychological studies of reading acquisition. The novel algorithm achieved state-of-the-art performance for line assignment of noisy gaze data. In my Psychology Bachelor’s Thesis, I continued to work with Sascha and built supervised classifiers to predict gaze in digital reading.

Data Privacy in Learning Management System Integrations #

In my final year project at the Hasso Plattner institute, I investigated data-privacy-compliant integration of third-party educational resources into online learning platforms using a pseudonymization approach under Jan Renz. Becoming intrigued with pedagogy for which the technology should serve, I wrote an article (in German) about didactics within e-learning environments in their blog: Dialogisches Lernen

Learning Apps #

InterestLearn #

Based on my own interest in language learning and frustration with limited topical choice for intermediate texts, I created InterestLearn, a language learning tool designed to engage students with personalized, proficiency-adjusted content and chatbot interaction including corrections. Check out the demo!

Podcastify AI #

At LauzHack 2024, my team ideated and implemented “Podcastify AI”, a personalized podcast generation app for WatchOS that allows language learners to listen to podcasts on their proficiency level. The unique thing about it: Learners can stop and select words they do not understand. After getting them explained in the context of the podcast, the word will reappear more often in the same podcast from this point on!

KnowTube #

KnowTube, created by the same team at LauzHack 2025, is designed as the virtual exit from mindless YouTube consumption to a more sustainable learning experience. It is a browser add-on that is active when videos are watched and embeds them into a learning environment that turns it into a more active learning experience with multiple choice questions, flash card creation and knowledge graph generation.