MUHAMMAD AHMAD MASOOD

Implemented representation learning and OOD domain generalization techniques, eliminating data annotation needs and cutting costs by 30%. Delivered a high-accuracy ANPR system processing 4 million+ vehicles daily.

Developed and deployed deep learning models in PyTorch for real-time object detection and tracking in traffic surveillance systems, optimizing inference for edge devices. Achieved a 35% reduction in false positives.

Developed a monocular depth estimation pipeline for low-light surveillance using self-supervised learning and transformer backbones, improving depth accuracy by 32%. Built a multi-view 3D reconstruction system for vehicle trajectory analysis.

Built and deployed 7+ machine learning applications across NLP and computer vision, including custom LLM fine-tuning and real-time video analytics tools. Fine-tuned LLaMA 2 and Falcon models for domain-specific text generation.

Gathered and preprocessed over 10,000 retinal images, applying data augmentation to reduce overfitting by 25%. Developed and fine-tuned a VGG16-based CNN achieving 92% accuracy and AUC of 0.95.

Pursued a Bachelor of Science in Computer Engineering, with coursework in data structures, algorithms, object-oriented programming, digital image processing, software engineering, AI system design, operating systems, and databases.