Our approach uses GANs for data gap reconstruction and a lightweight self-supervised MAE-Mamba framework for land-water segmentation to detect river erosion
Ongoing
TWIG uses segmentation masks in diffusion models to generate copyright-free images (IFIP Privacy 2024).
View Paper
TextDiffuser-RL optimizes text layouts for high-fidelity text-to-image synthesis using reinforcement learning.
View Paper
This study uses Sentinel-2 satellite imagery, remote sensing techniques, and machine learning to analyze riverbank changes in the Jamuna River Basin, Bangladesh.
View Project
Developed a deep learning model to analyze satellite images for identifying vegetation health and degradation patterns.
View Project
MangoScan classifies 10 Bangladeshi mango varieties using PyTorch and Streamlit with 95.05% accuracy.
View Project
BiLSTM model achieves 97% accuracy in token-level POS tagging on the POS 1 dataset.
View Project
A tech-driven social entrepreneurship solution addressing fire safety and disaster management in Bangladesh.Selected as a Bracu Hult Prize 2025 Semi-Finalist. View Project
Implemented an agent-based model to simulate and optimize evacuation routes and processes during cyclone threats.
View Project
Fully funded for tuition, accommodation, living expenses, insurance, and travel at Bard College, USA.
Japan Aerospace Exploration Agency (JAXA).
Among 84 teams.
Among 400 applicants.