Selected projects
A selection of projects spanning neural RF representations, GPS-denied geolocation, weather-aware networking, seismic early warning, and edge AI for safety-critical infrastructure.
A hardware-constrained neural field framework for actionable wireless channel modeling from sparse commodity WiFi CSI. WiNeRF embeds antenna geometry and phase uncertainty into the model, then reuses the learned field for RSSI coverage, beamforming, and AoA profiling.
Read the EWSN paper
I-NAV infers plausible locations from questionnaire-style navigation constraints in GPS-denied environments, matching commercial routing engines within 2–3 minutes of travel-time accuracy while remaining fully offline.
Using experiments on the NSF PAWR ARA testbed, I modeled how rainfall, snow, humidity, and pressure influence wireless latency and network KPIs for proactive network management.
At BUET-JIDPUS, I developed a self-supervised contrastive GNN for real-time seismic intensity prediction from short waveform windows, supporting AI-enhanced disaster response.
A Raspberry-Pi system for train detection, real-time alerts, and servo-controlled barriers at unmanned crossings, demonstrating low-cost computer vision for safety-critical infrastructure.