At Ogmen Robotics, I architected an edge-cloud multimodal pipeline for the Oro-pet care ecosystem, extracting more than 30 canine health signals. It powered proof-of-concept demos that helped secure four veterinary enterprise partnerships.
I also distilled 80K+ raw images into an 8K+ training set and used hardware-specific augmentations and hard negative mining to improve edge models. This reduced false positives by 78%, with a 2.5% drop in mAP50-95.
As a Computer Vision Engineer Intern at Ogmen Robotics, I designed a human-pet interaction module and a shared-compute ReID tracker. The tracker maintained targets through occlusions, reached a 25 Hz edge response rate, and reduced ID switches by 30–35%.
Earlier, as a Research Intern at Space Applications Center-ISRO, I tested laser spot distortion under atmospheric turbulence and compared centroiding algorithms. At ChiSquare X, I developed a 360-degree video-based mango yield estimation system that improved load estimation accuracy by 15%.
