This paper presents a methodology for generating personalized educational recommendations in the field of university-level mathematics, using Natural Language Processing techniques and Transformer-based models. First, an English-language dataset is preprocessed and translated into Spanish using the deep-translator library, aiming to adapt the system for a Spanish-speaking audience. Subsequently, a pretrained Transformer model is fine-tuned and tailored to the educational-mathematical domain. The quality of the generated content is evaluated using metrics such as accuracy and TransQuest, achieving an accuracy score of 0.96 and an average TransQuest score of 0.456. These results demonstrate that the system is effective in terms of coherence, relevance, and linguistic adequacy. This approach aims to promote autonomous learning, improve the understanding of mathematical concepts, and support personalized teaching in virtual or hybrid learning environments.