I am a final-year PhD candidate in Personal Health Informatics at Northeastern University in Boston, working at the intersection of human–computer interaction (HCI), ML/AI, and mobile health. My research centers on digital phenotyping—integrating mobile sensing with self-reported experiences to capture richer, more granular views of daily behaviors and mental states. I develop ML/AI-enabled, low-burden self-reporting methods that make intensive longitudinal data collection more user-friendly, sustainable, and accurate. I envision a future where intelligent digital phenotyping supports highly personalized, timely interventions that meaningfully enhance behavioral and mental well-being.
I am fortunate to be advised by Prof. Stephen Intille and to collaborate closely with Aditya Ponnada, Varun Mishra, Genevieve Dunton, Donald Hedeker, and Wei-Lin Wang. My work has been published in leading venues, including PACM IMWUT (UbiComp), Journal of Physical Activity and Health, JMIR mHealth and uHealth, Journal of Sleep Research, and Journal of Behavioral Medicine.
Outside of research, I enjoy reading, spending time outdoors, and playing board games.
📢 I am currently seeking postdoctoral and industry opportunities.
Oct, 2025 Presented our work on adaptive-length EMA and longitudinal engagement with μEMA at ACM Ubicomp 2025 in Espoo, Finland 🦌. Enjoyed my first Finnish sauna and cold sea-water dip ❄️.
Nov, 2024 Paper titled “Ask less, learn more: adapting ecological momentary assessment survey length by modeling question-answer information gain” accepted at Proceedings of the ACM IMWUT!
Oct, 2023 Received Distinguished Paper Award at UbiComp/ISWC 2023 🏆!
Mar, 2022 Paper titled “Contextual biases in microinteraction ecological momentary assessment (μEMA) non-response” accepted at Proceedings of the ACM IMWUT!
This project explores how large language models can support researchers in designing, specifying, and interpreting complex statistical models. The assistant helps users formulate research questions, choose appropriate model settings, and generate result interpretation aligned with best practices. It integrates with the MixWILD statistical software to offer contextual guidance, error detection, and step-by-step feedback. By combining domain knowledge with interactive AI support, the system lowers the barrier to advanced statistical modeling and boosts researchers’ accuracy and confidence.
This work evaluated μEMA’s longitudinal engagement over a year-long study. The results found that μEMA consistently achieved higher response rates and was perceived as less burdensome than traditional EMA, with the engagement gap especially pronounced among participants who disengaged from EMA. These findings highlight μEMA as a sustainable and scalable approach for intensive longitudinal data collection.
This study examines the feasibility of estimating the information gain of each EMA question to dynamically prioritize the presentation of the most informative questions while skipping less informative ones. When evaluated on four real-world datasets from three distinct EMA studies, the proposed method reduced imputation errors by 15%-52% compared to random question omission. This approach could enable more comprehensive construct measurement without increasing participant burden, or it could help reduce response burden to support more sustainable longitudinal EMA data collection.
Microinteraction EMA (μEMA), is a type of EMA where each survey is only one single question that can be answered with a glanceable microinteraction on a smartwatch. Prior work shows that even when μEMA interrupts far more frequently than smartphone-EMA, μEMA yields higher response rates with lower burden. We examined the contextual biases associated with non-response of μEMA prompts on a smartwatch. Based on prior work on EMA non-response and smartwatch use, we identified 10 potential contextual biases from three categories: temporal (time of the day, parts of waking day, day of the week, and days in study), device use (screen state, charging status, battery mode, and phone usage), and activity (wrist motion and location).
MixWILD is an interactive, open-source statistical software platform designed to make mixed-effects and location–scale modeling accessible to researchers working with intensive longitudinal data, especially the ones collected using ecological momentary assessments (EMA). I have contributed to expanding its capabilities and enhancing the interface to help users specify, fit, and interpret complex multilevel models without requiring deep statistical programming expertise. I also conduct ongoing interviews and usability testing with behavioral researchers to understand their needs, identify friction points, and inform new feature development.
This study explores the potential of using free open-source OpenStreetMap(OSM) geodatabases in semantic enrichment of individual’s frequented places. We created a three-level hiarchical taxonomy of place type labels, and a data processing pipeline from segmenting trajectory data to labeling visited places with OSM semantic tags. For a dataset with 93 people's 1-year location data, we successfully labeled 81.9% identified places and 60.3% of labeled places are matched with the groundtruth. The code base of curating OSM data and semantic enrichment will be open-source for other researchers to use.
