Research project · since 2026
LLM-assisted Geochemical Workflows
Investigating and verifying the correctness, reliability, and reproducibility of LLM-generated programs for geochemical data processing.
Overview
LLMs provide simplified access to advanced data analysis methods through programming languages such as Python. However, using LLM-generated programmes in a scientific context places significant demands on the programmes themselves. To draw meaningful conclusions and adhere to the principles of good scientific practice, data analysis must be accurate, reliable, robust, and reproducible.
However, non-specialists in computer science often cannot make such an assessment. Furthermore, LLMs themselves cannot guarantee the required levels of assurance. Consequently, as the use of LLM-supported data analysis in science increases, automated verification methods for LLM-generated programs are required.
We are currently conducting research on this topic within the team at the Chair of Reliable Software and Distributed Systems. We are developing and evaluating methods for verifying LLM-generated programs based on domain-specific criteria. In an initial pilot study, we developed a method that automatically examines LLM-generated programs to determine whether they generate REY plots that comply with geochemistry standards.
Results
We expect to be able to publish the initial results shortly. We will keep you informed.
Funding Information
This research is currently funded by the Research Initiatives and Support Programme (FORIS) at the University of Wuppertal in 2025.