My research focuses on information representation and computing infrastructure, and aims for AI for social good.​​

​​​​

• Research Showcase 1: World's first in-network machine learning platform for health monitoring, published here. This in-network ML infrastructure enables the future of ubiquitous health and patient monitoring, thereby increasing clinical capacity and reducing digital health disparities, especially in LMICs.

• Research Showcase 2: Large Language Models (LLMs) to Build Frontline Healthcare Worker Capacity in Rural India​, early-stage NeurIPs workshop paper here (main paper under peer review). Continued work on exploring the evaluation methods and the alignment of AI+GenAI computational linguistics with human language linguistics, as well as the biases introduced by human knowledge and user experience.

• Research Showcase 3: This RAEng DJT Research Fellowship project focused on developing machine-learning models and risk scores to monitor and predict gestational diabetes during pregnancy, a health condition that affects one in six pregnant women worldwide. Among these women, 50% will become type 2 diabetics within five years; it is paramount to identify the high-risk group to delay progression through earlier interventions.​​ Current work is to develop a health foundation model for multimodal time-series data and to explore the feasibility of using meta-learning with LLMs for explainable AI in health monitoring and disease discovery.

​​

​

Exploratory Research: 

 

1. AI and Healthcare: Learn from the Intellectual History of Technological Transformation

 

AI is not a binary force, especially in healthcare. As shown in Figure 1. Each maps to a recognisable type of frontier healthcare institution and has distinct implications for the physical, mental, and social dimensions of health. (1) The Extractive Model (liberal political economy): Knowledge flows one way—from workers and communities into AI systems. AI capability is hoarded at elite centres while the referring workforce is depleted and health inequity widened. This model is value-destroying—physically (reduced access at the periphery), mentally (deskilling, burnout, loss of professional agency), and socially (care communities fragmented, knowledge networks broken). Where extractive AI automates entry-level roles without creating new pathways, it risks collapsing healthcare career ladders for the next generation. (2) The Infrastructure Model (Chevalier’s conscious industrialism): This model is value-creating. Drolet’s analysis of Chevalier’s environmentally conscious political economy shows that sustainable infrastructure requires understanding humanity’s relationship to nature—directly relevant to climate-resilient healthcare AI. (3) The Circular Model (Leroux’s circulus): This model is value-regenerating: the circulus is not a metaphor—it is a measurable design principle for AI systems that regenerate the workforce’s capability rather than deplete it. The LMIC equity model—where AI must reduce inequality because workforce scarcity leaves no alternative. Leroux’s original vision was that what is taken from people must be returned to nourish complete humanity, not merely material productivity but physical, mental, and social flourishing.

​

​

​

​

​

​

​

​​​​

​

​

​

 

2. AI Diffusion in the Global Health Workforce

 

This project measures cross-country workforce dynamics and their temporal signal detection.

 

Primary results by LLM agents are shown here: OxHAEL Healthcare Occupation Crosswalk

​