Research

My research sits at the intersection of language, reasoning, and low-resource NLP.

Active projects

Language models for low-resource languages Building and training language models for Tulu and Sanskrit — two languages with rich literary traditions but limited digital representation. Focused on data collection, tokenization design, and transfer learning from related languages.

Sample-efficient physics reasoning Developing a model that reasons about physics problems from first principles rather than pattern-matching. The goal is strong performance with significantly less training data by grounding inference in physical law.

Diffusion models & reinforcement learning Reproducing and extending recent papers at the intersection of generative modeling and RL. Interested in how diffusion priors can improve exploration and planning in sequential decision-making.

Interests

• I rebuild architectures from scratch to understand them at a fundamental level
• Low-resource and morphologically rich languages
• Grounded, interpretable reasoning in neural models