Ravin Kumar

1. Developed ML solution for orange fruit grading using Intel RealSense with avg size error of 1.3mm (3mm allowed). 2. Built fruit tracking and multi-defects classification models with 93.44% accuracy. 3. Developed segmentation model for pomegranate defects quantification. 4. Reduced orange profiling time from 40 to 10 min. 5. Built data pipelines using AWS S3 using boto in python. 6. Deployed models using TorchServe on GCP. 7. Tech: PyTorch, TorchServe, MLOps, OpenCV, RealSense, LiDAR.

1. Developed an ML solution for MRI/X-ray reports in 54s (vs 1+ hr), detecting/quantifying Spine MRI & Chest XRay pathologies. 2. Created a Unet-based model for Central Canal/Lateral Recess prediction with 97.4% accuracy, 0.6mm error (1mm allowed). 3. Built object detection for Schmorl's Node (86.2% recall, 88% precision). 4. Trained/deployed segmentation for Thecal Sac on axial MRI (95.3% IOU). 5. Designed algorithm for multi-planar reconstruction across views. 6. Tools: Pytorch, MLFlow, Docker, Git etc.

1. Building cutting-edge products in the domain of Human-Computer Interaction Systems and Gesture Technology. 2. Developed models for the HCI and Gesture Technology, with inference on edge devices. 3. Our novel HCI, Gesture Technology was covered in the Thailand Tech Show 2020.

1. Developed the deep learning models, along with image processing methods of our "LYFLIKE" app for creating 3D avatars from images. 2. Built deep learning models, and used image processing approaches for detecting the cloth types from the images and generating 3D cloths with same textures using UV maps. 3. Development of AI pipelines, automation systems, and optimization via implementing parallelization techniques of multi-threading, multi-processing etc.