Better data, better models.

I am a full-time PhD scholar at BITS Pilani, Pilani Campus (since August 2026), working on machine learning for Earth observation. I am interested in how the quality of training data, such as noisy or inconsistent labels, shapes what models learn, and in measuring that effect instead of assuming it.

Before this, I completed my M.Tech in Computer Science and Engineering at B.K. Birla Institute of Engineering & Technology, Pilani (Bikaner Technical University). My dissertation, Deep Learning Paradigms for Multimodal Emotion Recognition, combined facial and speech signals and introduced a branch-preserving attention constraint to prevent attention collapse in multi-branch Swin models. Alongside it, I ran a statistical audit of label changes in the FER2013 dataset, with the methodology fixed before any results were seen.

I also mentor B.Tech students on their final-year projects.