Machine Learning Intern

What I worked on

• •Redesigned and fine-tuned two ResNet CNNs with CUDA acceleration for real-time blurry vs sharp image classification on Zebra devices, improving accuracy from 44% to 80% on close-up images and 72% to 80% on distant images.
• •Captured the source dataset myself and expanded it roughly fourfold through rotation, inversion, and mirroring augmentation to train on a usable sample size.
• •Ran human evaluation sessions with about 10 participants scoring image quality, creating the ground truth baseline used to benchmark and validate model performance.
• •Built a Python automation tool for dataset creation and evaluation loops that cut experiment turnaround sharply, and evaluated quantization for on-device deployment.