Researching Large Language Models for Scientific Discovery

I’m a 4th-year PhD student in Computer Science at the University of New Hampshire, working with Dr. Sam Carton. I build systems that leverage LLMs for materials science, healthcare, and education— with a focus on human-centered design & optimization.

Portrait of Satanu Ghosh

LLMs for Scientific Discovery

Post-training and prompting pipelines that ground LLM outputs in physics-based evaluations for inverse materials design and knowledge extraction from scientific literature.

Preference Tuning

Improving LM behavior using Bayesian & discrete optimization methods — shaping model outputs to satisfy domain-specific constraints that standard RLHF misses.

Expert-in-the-Loop AI

Active learning loops that embed domain-expert feedback directly into the training and evaluation cycle, improving reliability for high-stakes scientific and clinical applications.