My primary area of research is generative AI, CV in remote sensing and Medical imaging domain
Research Interests: Deep Learning for Computer Vision, 3D Reconstruction, Multi-Modal Models with applications in Remote Sensing, Medical Imaging, and Seismic Imaging
I work on representation learning and model evaluation, using internal representations as a practical lens into deep networks’ behavior. Before joining RVL, I worked in physics (nanostructure modeling) and computational neuroscience. I earned my B.Tech. from IIT Hyderabad in Engineering Science and Electrical Engineering, training that grounded me in rigorous mathematics and theory.
I work on turning satellite images into realistic 3D models of cities. I start from digital surface models (DSMs), which are 2.5D height maps of buildings reconstructed from multiview satellite imagery. DSMs capture the geometry of rooftops and facades but lack realistic textures or visual detail. My research uses 3D Gaussian splatting, a recent technique for representing 3D scenes as clouds of smooth primitives, to add high-quality textures directly onto these DSMs. This produces visually realistic, lightweight 3D city models that are useful for urban planning, simulation, and visualization.
My area of Research is in Continual Learning, especially in the domain of Generalized Category Discovery.
My PhD research explores schema-free entity relationship graphs for named entity recognition using large language models. The objective is to build agentic LLM-based systems that generalize beyond fixed ontologies, improving adaptability and performance across diverse and unseen domains.
My PhD research focuses on developing explainable AI systems specifically for fashion compatibility. The aim of my work is to move away from data-driven models that do not take nuances like style, culture, and material into account, towards models that actually cater to consumers by understanding these fashion undertones.
I research end-to-end and foundational models for autonomous driving, with a strong interest in computer vision and robotics to enhance perception, decision-making, and control in dynamic environments.
Primary research interest includes test and evaluation methodologies for neural networks with a specific focus of mechanistic interpretability for multimodal models.