At BIRDI, I work on LLM-guided RTL code generation. I built CompositeRTL, a dataset of verified designs whose v1 testbenches catch 92% of injected bugs, and improved model performance through fine-tuning and checkpoint merging.
I also built an agentic RTL benchmark and evaluation harnesses, and previously worked as an Intern at Samsung R&D Institute Bangladesh, where I performed software QA. At the University of Dhaka’s IIT, I teach short courses in Python, Microsoft Office, and practical AI tools.

