Voice-based AI for learning, training, and assessment
I'm Tony McGinn, a Senior Lecturer in Social Work at Ulster University. In 2024, I realised I could no longer rely on written essays to assess what my students actually knew. Too many submissions read like AI output. Besides that, my students are training to be social workers. They'll sit across from families in crisis, people in acute distress. They need to think on their feet and communicate under pressure. A polished essay tells me nothing about whether they can do that.
I started building HiEd.ai to solve this for my own classroom. The idea was: let students demonstrate their knowledge by talking, not typing. An AI agent conducts a realistic conversation, asks follow-up questions, and records everything for the lecturer to review and mark.
In December 2025, I tested it with 140 final-year social work students in a real, summative assessment. We surveyed the cohort afterwards: 68% preferred AI conversations to written assignments, and 78% preferred them over recorded presentations. Students ranked conversations first for building the skills they would actually use in practice. Several students said it was easier to talk to the AI than to a person, because it removed the social pressure of a face-to-face oral exam.
Since then the platform has hosted over 1,500 conversations across 400 students, at Ulster University and beyond, with trials underway at other UK universities. HiEd.ai has been adopted by Ulster Uni as spin-out enterprise and we are currently looking for partner Universities to trial, evaluate and help develop this platform into a world-class training and assessment platform which provides authentic assessments that correspond to working environments.
Figures from live use, and from an anonymous survey of the students who were assessed on it.
Now used by universities across the UK and Ireland, and by schools, not-for-profit organisations and emergency-service training teams.
About the survey: it was anonymous and went to the whole cohort. 69 students responded (49%), before they received their results. One result went the other way: 68% preferred multiple-choice tests to the conversation. The full findings are in our pre-print, now under peer review. Read the pre-print →
“I enjoyed marking the conversations and I can't say I have enjoyed marking more than a couple of essays in recent years due to students' use of AI.”
AI in Education Awards 2026, presented at the 6th National Conference on Generative AI in Education, Belfast
Papers directly related to this project:
McGinn, T., Pascoe, K.M. & Burns, P. (2024). Teaching social work students about evidence-based practice using peer-led learning and assessed conversations. Social Work Education, 44(6), 1519–1534.
Read on Taylor & Francis →(Pre-print) McGinn, T., Burns, P. & Crossan, B. (2026). Assessed conversations with an AI agent: social work students' views on an alternative to written assessment. DOI: 10.13140/RG.2.2.16223.21928
Read on ResearchGate →