At North Eastern Space Applications Centre (NESAC), I developed a UAV-based 3D reconstruction workflow combining COLMAP photogrammetry and Mip-NeRF 360 to reconstruct individual trees from RGB imagery. I reduced a 374-image dataset to 73 images through connectivity analysis and targeted selection, then validated the geometry against LiDAR with a 0.952 m MAE.
On my EchoJar project, I built a secure scheduled messaging platform with React.js and Django REST Framework. I improved backend performance by 40% through SQL query optimization, indexing, and database design, and automated expiration of 500+ messages daily with database triggers.
I also developed MediForecast, an ensemble model for hospital bed-demand forecasting, achieving 92% accuracy on 50,000+ patient records. I engineered predictive features and deployed the solution as a Flask application with an interactive dashboard.
For my explainable credit risk assessment project, I combined LightGBM with Temporal Convolutional Networks to model borrower data and macroeconomic trends. I used SHAP to produce interpretable risk predictions and built the project around feature engineering, preprocessing, and class-imbalance handling.
