Developed a fine-tuned YOLOv5-based ML pipeline for multi vehicle, emergency-vehicle, and accident detection, delivering 81% precision and 80% recall in evaluations— showing computer-vision accuracy and model-optimization expertise.
Developed an adaptive signal-optimization algorithm that dynamically reallocates green-light durations per lane using real-time vehicle counts and instant signal pre-emption for ambulances/fire trucks.
Implemented a PyTorch + OpenCV inference service with a PyQt5 desktop GUI, enabling real-time visualization of detections and intersection states; leveraged ThreadPoolExecutor for parallel video processing and Git for CI/CD —highlighting full-stack DevOps, multithreading, and cross-platform GUI development competencies.
