Supervisor at University of liberal arts Bangladesh
Research background is in embedded machine learning and biosignal processing, with a focus on making deep learning models run on low-cost, resource-constrained microcontrollers. Current work as a Research Assistant at Southeast University, Dhaka, under Dr. A. B. M. Sayeed Ud Doulah, covers wearable health monitoring, sensor fusion, and computer vision pipelines for clinical applications. Specific research interests include model quantization (INT8 and FP16), on-device personalization, and sensor fusion for wearable devices. Representative projects include TinySenseNet, a lightweight sEMG-IMU fusion network reaching 97.2 percent accuracy at 1.46 KB on an ESP32, and an on-device adaptive TinyML system for sEMG-based wheelchair control that raised accuracy from 56.8 percent to 95.5 percent after a 120-second on-device personalization step, deployed on a 256 KB Arduino Nano 33 BLE Sense Lite. Other work includes a quantized TabM model for glycemic event forecasting, compressed to 10.27 KB with a hypoglycemia recall of 96.0 percent, and a LoRaWAN-based emergency wearable for elderly care in low-connectivity settings. Beyond embedded systems, research interests extend to computer vision and behavioral signal analysis, including a side-view pose estimation pipeline for attention state detection in children with autism spectrum disorder, and a frame-level stochastic model of attention dynamics using a continuous-time Markov chain. Across these projects, the common thread is applying constrained, interpretable models to real clinical or assistive problems in low-resource settings, rather than optimizing purely for benchmark accuracy. This background positions supervision toward TinyML, embedded AI, and applied biosignal or health-sensing projects, where a defined hardware or dataset constraint gives students a concrete target to design around.