A custom object detection project where I trained a YOLOv5 model from scratch on a 5,000-image dataset to detect helmets, heads, and persons. Instead of relying on pretrained weights, I built the model ground-up to better understand performance dynamics — then used FiftyOne to visualize and evaluate predictions with precision. This project was inspired by a conversation at a computer vision meetup and reflects my growing focus on visual AI and safety-critical applications.