Developed a spatio-temporal risk assessment framework for the Franka Emika Panda robotic manipulator using Graph Neural Networks (STGAT, STGCN-LSTM). Built a processing pipeline for high-dimensional sensor data from ROS bag files to predict safety risks, improving human-robot interaction safety.

Research in appearance-based gaze estimation and redirection using Generative Adversarial Networks (GANs). Synthesized high-fidelity eye images on the MPIIGaze dataset and optimized training using a combination of perceptual, style, and content losses via VGG-16 features.

Multi-domain deep learning suite for medical diagnostic imaging and human behavioral analysis. Implemented CNNs (ResNet, EfficientNet) for diabetic retinopathy grading and temporal models (LSTM, GRU) for human activity recognition from smartphone sensor data. Leveraged PyTorch Lightning, Gin-config, and Ray Tune for training.

Designed and simulated a quantum image steganography pipeline. Demonstrated secure image encryption and decryption using quantum computing principles.