I completed my PhD at the Computer Vision and Pattern Recognition Unit, ISI Kolkata, working under the supervision of Dr. Utpal Garain. My PhD work has focused on trying to inculcate key features of symbolic AI techniques—such as domain constraint awareness and logical coherence—into deep learning-based systems.

I am currently exploring techniques to better understand large language model (LLM) reasoning and to promote constraint adherence in LLMs, with the goal of improving their reasoning, alignment, and overall reliability.

Recent Works

• sudoLLM: On Multi-role Alignment of Language Models

  Soumadeep Saha, A. Chaturvedi, J. Mahapatra, U. Garain
  Accepted to EMNLP (findings) | 2025

• KisMATH: Do LLMs Have Knowledge of Implicit Structures in Mathematical Reasoning?

  Soumadeep Saha, A. Chaturvedi, S. Saha, U. Garain, N. Asher
  Arxiv | 2025

• PhD Thesis: Domain Obedient Deep Learning

  Soumadeep Saha
  Indian Statistical Institute, Kolkata | 2025