• Collected and annotated over 12,000 indoor images; added 10 custom object classes and extended the COCO dataset with Indian-context categories for indoor mobility and navigation research.

• Trained a YOLOv7 object detection model in PyTorch, achieving 92% mAP and 91% average class-wise accuracy on complex indoor environments with diverse lighting and occlusion scenarios.

• Developed a real-time perception module using OpenCV for navigable path estimation, integrated into a mobile assistive app that also generates natural language scene descriptions for visually impaired individuals.

• Conducted a feature comparison study of Digital Asset Management (DAM) systems and created a detailed matrix to guide internal platform enhancements.

• Built a CNN-based image upload pipeline achieving 88% detection accuracy, enabling automated classification and reducing manual image processing overhead.

• Deployed a scalable ML pipeline using Docker and Kubernetes, reducing image retrieval time by 60% and saving over 80 hours weekly through visual clustering.