This master's thesis explores the use of large language models
(LLMs) for individualised support in secondary school
mathematical education. The theoretical part reviews the
capabilities and limitations of LLMs, with a focus
on their application in educational environments and
mathematics-related tasks. Based on the theoretical part, a
multiplatform mobile application enabling the generation of
mathematical exercises and tutoring dialogue using an LLM was
designed and implemented.

The application was then deployed in a real classroom setting
and evaluated using both quantitative and qualitative methods.
Although the quantitative results did not show a statistically
significant improvement in students’ academic performance, the
qualitative analysis provided insights into how students
interacted with the system and communicated with the LLM.

The work contributes to the understanding of how students
engage with LLM-based educational systems in mathematics
education and presents a functional prototype that may serve as
a basis for further development.