My research focuses on the intersection of AI/ML and hardware design, with an emphasis on applying automation (recently, driven mainly by LLMs) to hardware design and AI compilers.
Programming and optimizing hardware accelerators is notoriously difficult, requiring deep knowledge of both hardware and domain-specific languages (DSLs). This tutorial introduces LLMLift + Autocomp, two complementary frameworks that automate accelerator programming through LLM-driven compilation.
LLMLift demonstrates verified code lifting—translating general-purpose code into accelerator DSLs with machine-checkable correctness proofs. Autocomp shows how LLMs can drive automated optimization using structured prompting, hardware feedback, and iterative search to generate high-performance code that surpasses expert implementations across Gemmini, AWS Trainium, and NVIDIA GPUs.