Advisor: Prof. Nirmalya Roy. Trustworthy multimodal AI for contactless health monitoring.

• Develop contactless respiratory and vital-sign monitoring from RGB, NIR, and IR video, robust to motion, lighting, and occlusion.
• Study how vision-language models and health agents reason over physiological evidence, using calibration, evidence grounding, and counterfactual auditing to make them reliable.
• Deploy sensing and reasoning pipelines on edge hardware and mobile robots.
• Systems: RespFormer, E2RespUNet, RRPIPS, VLS-RR, CALIBRA, EviCal-RR, ViTaL-Diff.

In collaboration with the DEVCOM Army Research Laboratory (ARL).

• Built an on-device respiratory monitoring system for a Boston Dynamics Spot robot on NVIDIA Jetson Orin.
• Designed the vision and signal-processing pipeline for changing viewpoints, outdoor lighting, partial occlusion, and robot motion.
• Evaluated the system in field experiments at the ARL Robotics Research Collaboration Campus (R2C2), Graces Quarters, Maryland.

• Applied deep learning and quantitative ultrasound to obstetric imaging.
• Worked on image segmentation and registration for non-invasive tissue characterization and early screening.

HUMAP Fellow. Media feature

• Developed deep-learning methods for surgical video captured by wearable smart glasses: phase recognition, hand-frame extraction, and tool detection.

Field experiments and research demonstrations on the NSF-supported TRACE testbed for autonomous and cyber-physical systems. Gallery · Workshop

Demonstrated on-device contactless physiological sensing on a Boston Dynamics Spot robot at the ArtIAMAS University Experimentation Event, using the Distributed Virtual Proving Ground (DVPG). Event