Projects

CNN and logistic regression models compared across four blood-cell image classes.

Least squares, Ridge, and Lasso compared through cross-validation and held-out evaluation.

Hierarchical and classical planning for computer assembly, refurbishment, and shared-stock allocation.

Classical classifiers and recurrent neural networks compared on ECG signals.

A custom ResNet-50 detection prototype with polygon preprocessing and a Flask inference interface.

Swin–GPT-2 and ConvNeXtV2 hybrids for Nepali image captioning, evaluated on Flickr8k and Flickr30k.