TUM Legal Tech Group
The TUM Legal Tech Group develops AI systems that make law more accessible, transparent, and efficient. Our research combines natural language processing, machine learning, and computational legal reasoning to tackle real-world challenges in courts, legislatures, and public administration.
Our work spans:
- Legal NLP — processing, summarizing, and generating legal text, with a focus on integrating domain expertise into language models
- AI in legal advocacy and decision making — modeling judicial reasoning for European and German courts
- Digital-ready legislation — AI-supported legislative drafting and compliance automation
- Legal benchmarking — building evaluation frameworks to rigorously assess AI performance on legal tasks
We collaborate closely with federal and state ministries, courts, legal scholars, and legal practitioners to ensure our research translates into practice. Current projects include:
- Digital-Ready Tax Law — AI-supported legislative drafting for German tax law
- Generative Language Model for the Judiciary (GSJ) — developing generative AI to support judicial decision making
- TITAN — interdisciplinary research on robust AI deployment in the German justice system
- BenGER — a holistic benchmark and platform for evaluating LLM-based systems in German law
news
| Feb 25, 2026 | Two papers accepted at the upcoming ACM CS&LAW Conference in Berkeley (March 2026): Niklas Wais, Max Prior, and Runsheng Chen will present “AppeaLLM: Mimicking a Closed-World Environment for Court Decision Prediction,” and Matthias Grabmair co-authored “Thinking Longer, Not Always Smarter: Evaluating LLM Capabilities in Hierarchical Legal Reasoning” with Li Zhang, Morgan Gray, and Kevin Ashley. |
|---|---|
| Feb 2, 2026 | Congratulations to Santosh Tokala on a successful Ph.D. defense! He has moved on to Amazon, where he works as an Applied Scientist. |
| Dec 19, 2025 | Shanshan Xu successfully defended her Ph.D. dissertation! She is continuing her research as a Postdoc with the NLP North group at the University of Copenhagen. |
research supported by
