I'm a PhD student at Polytechnique Montréal and MILA (Québec AI Institute), advised by Dr. Amal Zouaq and Dr. Christopher J. Pal. My research focuses on semi-parametric approaches for eliciting complex reasoning from language models — exploring how non-parametric memory modules can extend what LLMs can reliably do.

A recurring theme in my work is offloading the burden of precise factual recall from parametric (model weights) to non-parametric (retrieval-based) memory. This line of thinking has led to work on reducing hallucinations in structured generation tasks like SPARQL query generation, and on improving geometric reasoning in vision-language models through modular code execution paired with retrieval-augmented function libraries.

Before my PhD, I was a research assistant at IIT Bombay working with Dr. Soumen Chakrabarti on temporal question answering. Prior to that, I spent several years as a graduate student at IISc Bangalore in Dr. Partha Talukdar's MALL Lab, where I worked on knowledge graph embeddings and deep reinforcement learning for NLP.