Servus, I’m Tom.
I'm a PhD researcher in Frank Hutter's group at the University of Freiburg. My research aims to automate the design and optimization of AI agents and multi-agent systems. I study how these systems can improve through changes to their prompts and structure, making them more capable and efficient while reducing the need for manual design.
About
My research focuses on agents’ prompts, how agents are connected, and how feedback on the overall system can guide improvements to individual agents. A central question is how to identify which parts of a system should change when we only observe whether the system as a whole succeeds or fails. Alongside this research, I contribute to promptolution, an open-source framework for prompt optimization, and am developing xmas, a library for multi-agent system optimization.
From 2020 to 2025, I worked at Airbus, first as a Dual Student and later as a Data Scientist. Across six internships in different departments and my subsequent role, I worked on applications ranging from predicting flight arrival times to developing similarity-search models for aircraft maintenance.
I completed my M.Sc. in Statistics and Data Science at LMU Munich, specializing in machine learning. During my Master's, I developed CAPO for cost-aware prompt optimization and investigated how positional encodings affect language models’ ability to use long contexts. My thesis received the department’s Best Master’s Thesis award.
Reach me at [email protected] or find me on GitHub, Google Scholar, LinkedIn, YouTube.
Papers
- CAPO: Cost-Aware Prompt Optimization AutoML 2025 · Methods Track
- promptolution: A Unified, Modular Framework for Prompt Optimization EACL 2026 · System Demonstrations
- Can Calibration of Positional Encodings Enhance Long Context Utilization? EACL 2026 · Findings
- CANTANTE: Optimizing Agentic Systems via Contrastive Credit Attribution Under review · 2026