Josef Zayan

I've built end-to-end computer vision systems for real-world retail automation, from custom dataset creation through real-time detection and checkout workflows.

For my Smart Shopping Cart capstone, I designed a YOLOv11, SAM2, and PaddleOCR pipeline that achieved 99.46% mAP@0.5 and 99.83% precision. I built and annotated an 18-class dataset of 4,632 retail product images and integrated the solution with a Flutter app, Flask backend, and Firebase database.

I've also benchmarked five CNN architectures for chest X-ray pneumonia detection, achieving 96.5% accuracy and 98.8% ROC-AUC with a custom CNN. I deployed the work as an interactive Gradio application for real-time model comparison and classification.

At FlyRank AI, I'm developing and deploying scalable machine learning pipelines in Python while researching and implementing AI models for production environments. I work with Python, PyTorch, TensorFlow, computer vision, model evaluation, and full-stack ML deployment.