Responsible generative AI in higher education: Expert validation of a compact RAG for technical and pedagogical robustness

Generative AI (GenAI) tools offer opportunities to support higher education students in adopting effective learning strategies, an area where guidance is lacking. Although evidence-based strategies such as retrieval practice and distributed practice enhance long-term learning, students frequently rely on less effective approaches due to limited instruction. GenAI could provide personalised, just-in-time support, but concerns about accuracy, trustworthiness, and pedagogical quality complicate its responsible use. This mixed-methods study examines how a compact retrieval-augmented generation system (<8B parameters) can support Dutch higher education students’ use of effective learning strategies. The system, grounded in curated domain knowledge, was evaluated using automated assessment (RAGAS) and expert validation. Seven domain experts evaluated fifteen responses using the Content Validity Index and qualitative feedback. While RAGAS indicated technical robustness, expert validation revealed insufficient pedagogical robustness in actionability and readability, demonstrating that automatic performance does not guarantee pedagogical robustness and underscoring the need for responsible GenAI in education.