At IDRBT, I designed, built, and deployed machine learning and deep learning pipelines for facial expression recognition, from data preprocessing through production integration. I also developed OpenCV automation pipelines that reduced manual annotation and preprocessing time by 40%.
In my projects, I implemented knowledge distillation with a ResNet50 teacher and lightweight ViT student, and built a hybrid GNN and Vision Transformer pipeline for facial expression recognition. I also designed DMSNet for mobile and edge deployment, with a real-time image processing pipeline achieving approximately 18 FPS on edge hardware.

