This site is currently under major renovation
Research statement
Whether an AI system in education can be trusted is an empirical, adoption-tested question. I answer it by building systems, deploying them at scale and measuring what happens.
Most of what I know about whether an AI system in education can be trusted, I learned by building enterprise systems that had to work for people who did not care about the technology, only the outcome. A decade of enterprise delivery, including a dental patient-management system across 200+ Hong Kong clinics and finance and hospitality platforms for enterprise clients, taught me that a system earns its place by surviving contact with real users under real constraints, not by scoring well on a benchmark. My research treats AI in education the same way, and everything below follows from that position.
1. Research identity: AI in education, tested by delivery
My work runs across three connected strands: software engineering education in transnational and offshore programmes, AI-enabled assessment and academic integrity, and the trust and governance questions AI raises in education. All three share a method rather than a topic: I build a system, deploy it at operational scale, and treat what actually happens when people use it as the evidence, rather than arguing from principle alone. Service on the IEEE Humanitarian Technologies Board and the IEEE Education Committee, and presentations to United Nations audiences, are a distribution channel for research findings, not a credential separate from them.
2. Deployed systems as the evidence base
The AI-powered peer-evaluation platform I designed is used by 300+ students a year at my own institution and has been adopted by five further universities. It cuts staff marking turnaround from roughly two weeks to three days and the student feedback cycle from two months to two weeks, and I extended it with automated content-safety moderation to test how it holds up under real submission volumes rather than curated test sets. The Dynamic Examination System introduced across three academic years, and the real-time programming test piloted at the Singapore Institute of Technology and later adopted by the Vocational Training Council and THEi Hong Kong, follow the same pattern: a system is only evidence once someone outside my own classroom has adopted it.
3. Methodological toolkit
My doctorate applied the Decomposed Theory of Planned Behaviour and Structural Equation Modelling to financial technology adoption in China’s Greater Bay Area. That gave me a quantitative, behavioural-modelling grounding I still use to study why people trust or resist a system. A full-time industry delivery career before I built peer-evaluation platforms gave me the ability to implement what the models suggest rather than hand it to someone else. A trust or adoption question gets modelled and tested behaviourally, then built and deployed as a working system, then measured again on real adoption data. My BSc, with a minor in Engineering Management, adds the systems-engineering half of that loop: it is the reason I design deployments as instrumented experiments from the outset, rather than adding measurement after a system is already live.
4. A forward research agenda
Three directions extend directly from this record. First, a formal, SEM-based study of trust and adoption determinants in AI-assisted assessment across student cohorts, using the peer-evaluation platform’s existing multi-university deployment as the field site and extending my doctoral methodology into a domain it has not yet been applied to. Second, extending my transnational software engineering education research into a study of how AI-enabled assessment and integrity tooling transfer, or fail to transfer, across institutions, jurisdictions and student populations, targeting the venues where computing education and AI already overlap: ACM and IEEE journals and flagship conferences in software engineering education. Third, translating the trust and governance work into deployable fairness and content-safety tooling for higher education more broadly, moving that conversation into shipped infrastructure the way the peer-evaluation platform did for assessment.
5. Interdisciplinary and applied collaboration
Applied AI work with medicine, nursing and law follows the same build-then-evaluate method: GenDatr, a visualisation tool for probabilistic data generation in medical data science; an AI chatbot for a nursing ethics and law module; AI-enabled IoT patient-simulator work across medicine, nursing, engineering and computing; and the WellBeing Together community programme with elderly participants. I read a research-funding criterion as asking whether a candidate can win resources, hold a budget and deliver against milestones. I hold one competitive academic award, as Co-Investigator on a Learning and Teaching Development Fund project, and in quantity I hold funded project delivery with budget, milestone and risk accountability: an unmanned smart-shop laboratory (Project Manager), a logistics management system, and a supply chain digital twin delivered with an external engineering partner and two public transport agencies.
6. Supervision and a student-collaborated research group
I co-supervise higher degree by research candidates, and I supervise undergraduate and graduate capstone projects, coordinating tutors, assessors and external stakeholders. My research group is built around student collaboration by design: several of my publications list a student as first author with me as supervising co-author, and I treat a supervised project’s publication, not merely its completion, as the standard I hold students to. Students are research partners in how AI-enabled tools get designed, deployed and evaluated in their own classrooms, not merely subjects the research is done on. Growing a research group means extending that partnership deliberately, recruiting research students into the strands above and pairing each with a deployed system or an external partner, so a thesis is answerable to more than a committee.
In summary
I want a research programme in AI and education that is judged the way I judge my own systems: by whether it is adopted, whether it holds up under real institutional constraints, and whether it earns the trust of the people who use it.
Papers, DOIs and BibTeX: research.peteryau.com. See also the research overview and the teaching philosophy.
