Efficient Tax Administration through Artificial Intelligence
The Bavarian AI Taxation Laboratory (BAI) is a joint research and development initiative by the University of Technology Nuremberg and the Bavarian State Office for Taxation. The goal: to bring innovative AI solutions directly into practice – with real benefits for citizens.
A key focus is the development of a fine-tuned large language model that provides clear, accessible answers to tax-related questions and delivers legally sound reasoning. Automated systems are also being designed to detect problematic areas in tax returns, identify errors more efficiently, and reduce processing times.
The result: streamlined administrative processes, lower costs, and better decisions for a modern tax system and improved public service.
The collaboration between UTN and the Bavarian State Office for Taxation is currently in development and growing steadily. At the center of this effort is an interdisciplinary team dedicated to exploring the use of large language models in the tax domain and creating new AI-driven solutions. The lab is led by Prof. Dr. Josif Grabocka, Professor of Machine Learning at UTN, and is currently based at UTN’s interim facility on Ulmenstraße in Nuremberg.
News
February 26/27, 2026
The Bavarian AI Tax Lab (BAI), led by Prof. Dr. Josif Grabocka, is a joint research initiative of the University of Technology Nuremberg (UTN) and the Bavarian State Tax Office. Its focus is on researching and developing AI solutions for taxation. On February 26 and 27, BAI held its first know-how transfer symposium. The thematic focus was on best practices for integrating retrieval-augmented generation (RAG) into large language models. The event brought together numerous experts from various interested public institutions and provided space for professional exchange, knowledge transfer, and discussion of specific potential applications.
January 10, 2025
The signing of the agreement between the University of Technology Nuremberg and the Bavarian State Office for Taxation marks the official start of their collaboration.
Publication
Guri Zabërgja, Rafiq Kamel, Arlind Kadra, Christian M. M. Frey, Josif Grabocka
End-to-End Compression for Tabular Foundation Models
International Conference on Machine Learning (ICML 2026)
https://arxiv.org/pdf/2602.05649
