vibe-coded, already useful still vibe-coded, but seems useful already - proper pass when I find the time
Tom Zehle
tomzehle.com

Tom Zehle

PhD Researcher, Machine Learning

University of Freiburg / ELLIS Institute Tübingen · Research group of Frank Hutter

Freiburg, Germany · [email protected] · GitHub · Google Scholar · LinkedIn · YouTube

PhD researcher working on prompt optimization and multi-agent system optimization. Research vision: automated design and optimization of multi-agent systems (“AutoML for Agents”). Task lead in the EU ELLIOT project (WP3). Previously data scientist at Airbus; M.Sc. in Statistics and Data Science from LMU Munich. First author of four public papers (one sole-authored); maintainer of the open-source framework promptolution.

Download PDF

Tom Zehle in front of the Siegestor in Munich

Experience

PhD Researcher · University of Freiburg / ELLIS Institute Tübingen

Oct 2025 – present · Freiburg, Germany

  • Research on prompt optimization and multi-agent system optimization.
  • Task lead in the EU ELLIOT project (WP3, Task 3.3: fine-tuning of MGFMs).
  • Author of five publications to date (four first-author, one sole-authored); see Publications below.
  • Supervising B.Sc./M.Sc. students and a student assistant; see Supervision below.
  • Reviewer for NeurIPS 2026 and Nature (2025).

Data Scientist · Airbus

2023 – 2025 · Munich, Germany

  • Engineered a similarity-search model for the Eurofighter support center; now in production, used by system engineers to accelerate incident handling.
  • Built fine-tuning scripts, a comprehensive evaluation framework, and production backend services and APIs for model deployment.
  • Developed a prompt-optimization proof of concept and an AI-powered export-classification pipeline for sensitive documents.

Dual Student, Data Science · Airbus

2020 – 2023 · Munich / Ingolstadt / Getafe

  • Practical phases across computer vision, traditional ML, and NLP.
  • Built a transformer-tabular hybrid model to predict flight arrival times.
  • Developed a computer-vision algorithm for the ExoMars rover project and a scene-change-detection system for production safety.

Education

M.Sc. Statistics and Data Science · LMU Munich

2023 – 2025

Machine Learning track. Grade: 1.1.

Thesis (1.0): “Understanding Temporal Structure in Text – A Study of Positional Encodings in LLMs”; published as the positional-encodings calibration paper at Findings of EACL 2026.

B.Sc. Business Informatics (Data Science Specialization) · DHBW Ravensburg

2020 – 2023

Dual study with Airbus. Grade: 1.4.

Thesis (1.0): fine-tuning transformers for similarity search across aircraft documentation and incident reports.

Publications

Supervision

Awards & Media

Award
Airbus “Fly Your Ideas” Competition 2024, with L. Backhaus and N. Lethaus [link]
Podcast
“How does an Airbus data scientist keep up to date?”, Industrial AI Podcast, May 2025 [listen]
Videos
Paper walkthroughs and demos, e.g. the promptolution EACL 2026 demonstration [channel]

Skills

ML research
Prompt optimization, multi-agent systems, AutoML, LLM evaluation
Engineering
Python, PyTorch, open-source development (promptolution maintainer), agentic coding workflows
Statistics
Experimental design, multi-objective optimization
Languages
German (native), English (fluent), Russian (intermediate), Spanish (basic)

This page is print-friendly — use your browser’s “Print / Save as PDF” for a downloadable copy.