Detail
Abstract
Traditional educator-based grading is often limited by delayed feedback, subjectivity, grading drift due to fatigue, and a lack of personalization in feedback. Large language models (LLMs) can improve the efficiency and consistency of grading by providing instant feedback before students commence a new topic, applying rubrics and model answers consistently, generating personalized guidance for students, and summarizing whole-class performance to support teaching. To address this, we present Vox-LM, a multimodal AI tool designed to grade individual and batch short-answer questions (SAQs) across disciplines. Vox-LM is a prompt-engineered system built on the Qwen 3 vision-language instruction-tuned model, which evaluates student responses against model answers and rubrics, with or without the addition of human-marked exemplars. The system outputs student marks, highlighted summaries showing correct, incorrect, and uncertain phrases in student responses, rationales for the marks awarded, personalized feedback, Vox-LM’s confidence in the assigned grades, and norm-referenced comparisons for batch grading. Preliminary evaluation of Vox-LM suggests that its grading achieved a high positive correlation with marks assigned by a senior examiner. To support real-world educational use, Vox-LM also includes a Human Override module for teacher verification, a Student Challenge module for student re-marking requests with justification, and a class summarization module to help teachers identify learning gaps and inform subsequent learning activities.
About the Speaker(s)
Prof. John Adeoye is an Assistant Professor in Digital and Precision Dentistry at the Faculty of Dentistry, University of Hong Kong (HKU), where he develops AI-based prediction systems for head and neck oncology. With over 40 peer-reviewed publications, he has pioneered data-driven platforms for oral cancer decision support that significantly outperform traditional methods. To prepare the next generation of dentists and clinicians, Adeoye co-developed the AI Literacy in Dentistry curriculum for second year undergraduate dental students. His current pedagogical research focuses on developing GenAI models for automated assessment and feedback generation, as well as seeking to integrate these tools into existing learning management systems at HKU to enhance student learning.
Prof. Michael Botelho is the Assistant Dean of Wellness at the Faculty of Dentistry, HKU and has been involved with educational research, reform and innovation for over 30 years. He is an Associate Editor of the European Journal of Dental Education and Past President of the Educational Research Group of the IADR and past board member of the Association of Dental Educators of Europe. He has over 100 peer-reviewed research papers with over 25 of these in dental education and over 15 educational grants as PI. He has a number of education awards including the prestigious Tertiary sector Hong Kong UGC Teaching Award and University of Hong Kong University Distinguished Teaching Award. He developed Video Vox 1 and Vox 2.0 learning management system at HKU and is updating its functionality with ongoing TDLEG grants in portfolio design.
Ms. Trinity Jiao is a Project Manager at the University of Hong Kong, where she leads the development of strategic educational initiatives. She has overseen the end-to-end delivery of Hong Kong's first in-house Learning Management System and a university-wide student wellness platform–both designed to enhance learning experiences and support student well-being.
She works closely with faculty, academic leaders, and students to translate pedagogical goals into practical, scalable projects. Trinity also helped establish a "Students as Partners" community of practice, growing it to over 100 core members and engaging more than 400 participants in collaborative learning and curriculum dialogue. She is passionate about bridging project management, educational technology, and student voice to create meaningful impact in higher education.