research
From everyday data to clinical evidence.
I study how to take AI for health from everyday data to clinical evidence. My work follows one research cycle, in which each round of trials and deployment produces the data for the next:
1
Sense
Digital phenotyping from wearables, app diaries, and social media
2
Model
AI algorithms and conversational agents for assessment and intervention
3
Evaluate
Benchmarks and evidence synthesis before agents reach people
4
Validate
Randomized trials of real-world effectiveness
↺ Trial and deployment data feed the next round of sensing.
Research interests
- Digital phenotyping with wearables and smartphone diaries
- Conversational AI agents for behavior change and digital therapy
- Evaluation of large language models and AI agents in health
- Randomized controlled trials of digital health interventions
- Digital sleep medicine and adolescent mental health
- Vaccine confidence and health misinformation
The cycle in practice
| Stage | Vaccine communication (HKU, completed) | Sleep and mental health (CUHK, ongoing) |
|---|---|---|
| Sense | Social listening on public perspectives toward mRNA vaccines and health AI | SmartSleep for a 6,000-adolescent cohort, 乐眠无忧, and Actigraph, Huawei, and Samsung wearables |
| Model | Multilingual vaccine chatbots and an AI fact-checking extension | AI agents delivering motivational interviewing, imagery rehearsal therapy, and CBT-I; sleep–wake and sleep-stage classification; prediction of adherence and remission in self-help digital CBT-I |
| Evaluate | Benchmark of large language models on multilingual vaccine knowledge | Systematic review of AI agents for mental health; psychiatric reasoning in small language models |
| Validate | Randomized trials in China (Nature Medicine), Japan (JAMA Network Open), and Thailand, Hong Kong, and Singapore (npj Digital Medicine), plus a randomized experiment of the fact-checking extension | Randomized controlled trial of the motivational interviewing chatbot |