About Me

I am an AI for Science researcher with extensive experience (5+ years) in uncertainty quantification in AI and designing comprehensive evaluation benchmarks to measure predictive accuracy, calibration and out-of-distribution robustness of AI models for scientific tasks. I am interested in bridging the gap between the theoretical and empirical sides of AI for Science to develop statistically rigorous AI systems.

I recently completed my PhD at the Department of Physics and Astronomy, University of Manchester. My thesis, Towards robust and reliable representations in astronomy, deals with the challenge of designing trustworthy deep learning models for morphological analysis of radio galaxies. Previously, I completed a Masters by Research (MRes) in Astronomy and Astrophysics at the University of Manchester and a Bachelors of Technology (B.Tech) in Electronics and Communications Engineering from Guru Gobind Singh Indraprastha University, New Delhi.

Through my research, I have investigated the inherent trade-off between accuracy and calibration, explored how optimisers induce inductive biases which can affect distributional robustness, and established links between model uncertainty and data geometry to identify failure modes in self-supervised foundation models for astrophysics. I have also translated theoretical insights from modern statistical frameworks like PAC-Bayes theory and information geometry for improving my variational inference models.