I am a Ph.D. student (joined in January-2021) in the Department of Computer Science at the Indian Institute of Technology Kanpur (IIT Kanpur) where I work under the guidance of Dr. Ashutosh Modi.

Previously, I worked with Hyundai Mobis R&D (Mobis Technical Center of India) as a Senior Research Engineer. I was a part of the Computer Vision team of the ADAS (Advanced Driver-Assistance System) Department, where I majorly worked on vision algorithms for rear-view cameras and generative models for synthetic scene generation.

I am interested in how reasoning takes place in Language Models and where their explanations/interpretations succeed or fail, and what their internal structure (representational or weight subspace) reveals about decision-making.

Selected Writings

• Geometry of Decision Making in Language Models (NeurIPS 2025): Explores how decision-making points emerge from internal representational structure and what this implies for explanation.
• Beyond Components: Singular Vector–Based Interpretability of Transformer Circuits (NeurIPS 2025): Argues that component-level explanations (widely used in mech interp) may be insufficient for understanding transformer behavior and proposes abstract subspace analysis in weight space.
• Calibration Across Layers: Understanding Calibration Evolution in Language Models (EMNLP 2025): Investigates how confidence and reliability evolve internally rather than appearing only at the output level.
• Towards Quantifying Commonsense Reasoning with Mechanistic Insights (NAACL 2025): Examines whether commonsense reasoning can be rigorously evaluated and meaningfully localized within a system’s internal mechanisms.
• COLD: Causal Reasoning in Daily Activities (NeurIPS 2024): Introduces a framework for modeling cause-and-effect relationships in everyday tasks (helping link actions, consequences, closed context), and formulates how causal commonsense reasoning can be investigated through interventions, counterfactuals, and backdoor criteria.