Assigning each undergraduate student a thematically aligned thesis advisor is labour-intensive and prone to ad-hoc decisions. This study introduces a fully automated, content-centric pipeline that profiles faculty expertise by mining 9278 Spanish-language theses from the Escuela Politécnica Nacional repository. PDF text is cleaned, embedded with multilingual Sentence-BERT, clustered via BERTopic, and summarized through class-based TF–IDF to yield advisor–topic relevance scores. Twelve strategic research themes were used as test queries. For every theme, the system produces a ranked shortlist of supervisors; between 5% (Data Analysis) and 42% (Network Security) of advisors surpass the relevance threshold \(\tau =0.5\) , signaling a strong topical affinity. Validation with twenty-six master’s students shows a top-1 advisor match of 55% and Recall@10 of 82%, vastly outperforming random selection and reducing search time from days to minutes. This robust, replicable methodology significantly streamlines academic management, enhances transparency, and provides a foundation for future multilingual and dynamic topic-modeling applications.