LLM-Based Interpretation of Student Talent-Interest Profiles for Personalized Career Guidance

Student talent-interest assessment can support educational and career guidance, but its practical use requires more than reporting raw scores. This study proposes a retrieval-grounded LLM-based framework for interpreting Talent-Interest Assessment (ABM) results into individualized explanatory narratives and career-oriented recommendations. Using 614 usable student records, the framework combined structured student profiles with two vector stores containing the official ABM guideline as the primary interpretive source and supporting career-related references as contextual enrichment. The generated outputs were evaluated using a rubric-based LLM-as-Judge framework as an automated quality-estimation mechanism. The results show that the highest mean scores were obtained for clarity and organization (4.331) and supporting-talent interpretation (4.000), while the overall score reached 3.774. Operational analysis further showed that the retrieval-generation and LLM-as-Judge evaluation stages consumed 6,382,552 tokens in total, with an estimated project-level cost of USD 18.08, or approximately USD 0.0294 per student. These findings indicate that retrieval-grounded LLMs can provide a practical approach for transforming multidimensional student assessment results into personalized and interpretable guidance narratives, while also highlighting the need for human validation, retrieval-quality analysis, and token-efficient designs for larger-scale deployment.