Developed my master's thesis on mitigating spurious correlations in image datasets via counterfactual image augmentation, under the supervision of Prof. Hassan Khotanlou.

Research on mechanistic interpretability of LLMs, resulting in a publication at ACL 2026, under the supervision of Dr. Mahdieh Soleymani Baghshah.

Investigated and implemented culturally-aligned LLM post-training approaches by creating specialized datasets and leveraging Multiple LoRA Experts, resulting in the model Matina and a published paper at ACL 2025.