About

Hi, I’m Frederic. I’m a first-year PhD student in Computer Science at the Ludwig-Maximilians-Universität in Munich, where I work on interpretable language models with Professor Hinrich Schütze at the Center for Information and Language Processing. My position is fully funded through the PhD Matchmaking Program of the Munich Center for Machine Learning (MCML).

Before my PhD, I completed my master’s in Computer Science at the Hasso Plattner Institute. My master’s thesis, supervised by Dr. Michal Štefánik, Konstantin Dobler, and Professor Gerard de Melo, looks at decreasing the prompt sensitivity of language models and was accepted to the Findings of EMNLP 2026. I carried out part of the research while visiting the LLMC-Group of the National Institute of Informatics in Tokyo.

During my master’s, I also completed a research internship in the Research & Innovation team at SAP in Palo Alto, where I worked with Professor Diyi Yang from Stanford University on the collaboration of coding agents. The resulting benchmark, CooperBench, was accepted to COLM 2026.

I am interested in making large language models more robust and efficient, particularly by exploring how information encoded in their latent representations can improve training and generalization. In this area, I have looked into:

• Prompt-robust LLMs
• Soft prompts
• Hierarchical LLMs
• Efficient pre-training methods

I have also looked into:

• Collaboration of agents on code tasks
• Reinforcement Learning for Code LLMs

You can find my publications here and my other projects here.