Xin Liu: “Teaching AI to Understand Your Health”

View at publisher Summarise with VIRAT
Authors: Lakmal Meegahapola
Cited by: 0
References: 7

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Xin Liu: My research focuses on three interconnected directions aimed at making personalized health intelligence accessible to everyone. First, I build personal health agents that reason over wearable sensor data. Our work published in Nature Medicine showed that large language models (LLMs) grounded in wearable data can generate expert-quality sleep and fitness coaching, and our agentic system demonstrated how LLM agents can use tools and multistep reasoning to transform more complex health queries into actionable health insights. Most recently, we designed a multiagent personal health system that orchestrates data analysis, domain expert reasoning, and behavioral coaching into a unified framework. Second, I develop wearable foundation models trained on large-scale sensor data to perform diverse health tasks with a single model. Through projects like large sensor model (LSM) and sensor-language model (SensorLM), we created models that learn general representations from wearable time-series, enabling few-shot or zero-shot health understanding without task-specific training. Third, I contribute to advancing LLM reasoning capabilities as a core contributor to Google’s Gemini 2.5, focusing on tabular and data science reasoning, work that directly strengthens the analytical backbone of health agents.

Fields of Study & Subjects

Context-Aware Activity Recognition Systems Artificial Intelligence in Healthcare and Education Mobile Health and mHealth Applications

Publication details

Journal / source
IEEE Pervasive Computing
Volume
25
Issue
1
Pages
72–74
Publication date
1 January 2026
Language
EN