Résumé : One important step in software development is testing the finished product with actual users. These tests aim, among other goals, at determining unintuitive behavior of the software as it is presented to the end-user. Moreover, they aim to determine inconsistencies in the user-facing interface. They provide valuable feedback for the development of the software, but are time-intensive to conduct. In this work, we present GERALLT, a system that uses Large Language Models (LLMs) to perform exploratory tests of the Graphical User Interface (GUI) of a real-life engineering software. GERALLT automatically generates a list of potential unintuitive and inconsistent parts of the interface. We present the architecture of GERALLT and evaluate it on a real-world use case of the engineering software, which has been extensively tested by developers and users. Our results show that GERALLT is able to determine issues with the interface that support the software development team in future development of the software.