A selection of projects spanning wireless networking, sensing, and edge AI. The highlights below include real-world deployments and data-driven methods.
Reliable low-latency wireless is crucial for real-time applications. We studied how rainfall, snow, humidity, and pressure influence latency using experiments on the NSF PAWR ARA testbed and weather data from ARA's weather stations. We built predictive models that relate weather to network KPIs, enabling proactive, weather-aware network management. I was invited as a guest-speaker to present this work at NSF MERIF and ARAFest.
As a research engineer at the Japan Institute of Disaster Prevention and Urban Safety, I developed a self-supervised contrastive GNN that predicts seismic intensity in real time from short waveform windows. The model achieved high accuracy (low MSE on EMS-98) and supports alerts several seconds before strong shaking—contributing to AI-enhanced disaster response.
For improving rail safety at unmanned crossings, I built a Raspberry-Pi system with computer vision for train detection and servo-controlled barriers. The prototype performs on-device inference and triggers alarms and gate control in real time, demonstrating low-cost edge AI for safety-critical infrastructure.