Eeham Khan

I'm a PhD student working with Dr. Leila Kosseim and Dr. Ali Ayub on making large language models more robust and adaptable. Specifically, how they can learn from new domains without forgetting what they already know.

Recently, I've been working on adapting LLMs to Quebec French using parameter-efficient fine-tuning, as well as exploring novel methods to tackle class-incremental learning challenges in both computer vision and NLP. We released the first Quebec French LLMs on Hugging Face, and I'm excited about the potential for low-resource dialect adaptation more broadly. I also spent some time at CRIM building RAG pipelines with explicit reasoning chains and trustworthiness modules for domain-specific applications.

Beyond research, I really enjoy teaching. I've TA'd several courses such as Statistical NLP, Machine Learning, and AI, and I mentored student projects through the Google Developer Student Club at Concordia.