Motivation: The three-dimensional Geometry Understanding of Large Language Models is a current research topic with great potential and few published approaches. In this thesis, the applicability of research results from PLCM is to be investigated.

Task: First, the state of the art in Geometry Understanding of LLMs will be assessed. The primary focus will be on investigating methods that process three-dimensional inputs and are not just tracing back to image processing algorithms.

Subsequently, a selection of these methods will be tested for the description of class-dependent features of a classifiable data set.

Then, it will be investigated if and how these methods can be transferred to mechanical designs. Of particular interest here is the transfer of existing geometry encoding approaches developed at PLCM to the latent spaces of language models.