Arman Kareyan

I'm building end-to-end machine learning pipelines in Python and PyTorch, with a focus on computer vision, deep learning, and NLP architectures.

For diabetic retinopathy classification using the APTOS 2019 dataset, I architected a pipeline to detect five stages of disease from fundus imagery. I fine-tuned a ResNet-50 backbone with custom linear and dropout layers to improve generalization.

I addressed class imbalance through stratified sampling and image augmentation, then implemented Cross-Entropy Loss, Adam optimization, learning-rate scheduling, and macro-F1-based early stopping. I also integrated pytorch_grad_cam to validate that the model focused on pathology regions.

My training includes advanced Python and OOP, SQL and SQLite persistence, data analysis with NumPy and pandas, classical machine learning, CNNs, object detection, segmentation, generative AI, and Transformer architectures.