Three areas of research — accessible AI-driven learning technologies, AI across the lifespan, and participatory & responsible AI design — and the projects running under each.
We design, build, and rigorously evaluate AI-powered tools that make learning accessible to students with disabilities from the ground up, not as an afterthought. This includes course-specific AI tutors that support instructors across disciplines in scaffolding student learning responsibly, accessible learning management systems co-designed with blind and low-vision teachers and students, and evaluations of how well existing generative AI tools align with accessibility standards and Universal Design for Learning principles.
Across this work, we pair usability testing with the lived perspectives of the students and teachers who use these systems daily, ensuring that accessibility is measured not just against technical standards, but against real classroom experience.
Education doesn't end at graduation, and neither does our research. We study how AI-driven technologies — including conversational chatbots, literacy tools, and interface design choices — can support learning, independence, and trust for populations often overlooked in mainstream AI research and design, with a particular focus on older adults.
Our work has examined how different chatbot interfaces (text, voice, and avatar-based) shape older adults' trust and usability perceptions, and how hands-on AI literacy interventions can help older adults navigate the benefits and risks of generative AI with greater confidence.
This line of research reflects our belief that meaningful access to AI shouldn't diminish with age, and that lifelong learners deserve technologies designed with their needs, not just their limitations, in mind.
Our research is grounded in the conviction that the people most affected by a technology should help shape it. We use participatory and co-design methodologies throughout our work, from workshops with blind and visually impaired teachers and students to interviews with instructors navigating the ethical complexities of AI tutors in higher education.
Alongside this human-centered design process, we examine what responsible AI adoption actually requires in practice: transparency about system limitations, human-in-the-loop safeguards, data governance that protects vulnerable users, and guardrails that support learning rather than replace it. Rather than treating ethics as a checklist applied after development, we investigate how responsible design principles can be embedded into AI systems from their earliest design stages.
Active projects in the lab, grouped by the research area each belongs to.
Despite rapid investment in generative AI across higher education, how instructors create, evaluate, and implement course-specific AI tutors — and how students actually use them — has not been thoroughly studied. We explore how instructors develop course-specific tutors and judge their effectiveness, and where student use aligns, or fails to align, with what instructors expect.
As generative AI becomes ubiquitous, it is crucial to equip users — particularly vulnerable populations such as older adults (65+) — with the knowledge to weigh its benefits and risks. Older adults often hold greater reservations about adopting emerging technology and need tailored literacy support. We built Litti, a GenAI literacy chatbot that introduces older adults to generative AI and lets them practise tasks with it, then evaluated how the experience shaped their understanding of what the technology can and cannot do.
Conversational agents increasingly support older adults through text, voice, and embodied interaction. Modality is often treated as a presentation choice, yet little is known about how it shapes older adults' perceptions of an agent. We compare text-based, phone-call-based, and graphical avatar-based interfaces to see how each affects perceived ease of use, accessibility, and reliability.
Large language models are increasingly embedded in constraint-heavy conversational interfaces that carry out state-changing tasks. While they enable flexible dialogue, their probabilistic behaviour introduces breakdowns: semantic misalignment, execution-state corruption, and pragmatic failure. Little is systematically understood about how these manifest in correctness-critical domains, or how architectural design shapes recovery. We examine ways to evaluate breakdown and repair behaviour in conversational user interfaces.
The lab is methodologically mixed, but participatory and co-design methods anchor it. A single project might move from co-design sessions with learners and educators, through building a working prototype, to a study of what that prototype actually changes for the people using it.
We welcome prospective graduate students, undergraduate researchers, and collaborators who care about access as much as they care about the technology.
