AI-based exam question generation system for UBTEB

Abstract

The increasing demand for scalable and efficient examination systems in technical and vocational education has highlighted the limitations of traditional manual question-setting processes. This project presents the design and development of an AI-Based Exam Question Generation System tailored for the Uganda Business and Technical Examinations Board (UBTEB). The system leverages Natural Language Processing (NLP) techniques using the T5 Valhalla transformer model and incorporates a lightweight Retrieval method. It enables automated generation of high-quality, curriculum-aligned examination questions from both structured (curriculum documents) and unstructured (examiner input) content. The system is designed with offline capability to ensure security and usability in resource-constrained environments. Evaluation results demonstrated that the system produced syntactically correct, semantically relevant, and non-repetitive questions across various subjects, achieving a BLEU score of 0.371 and high approval ratings from domain experts. The project contributes a practical, scalable, and secure solution for modernizing assessment in Uganda’s technical education sector and sets a foundation for broader applications in educational tools.