"ArchNative team on AMD Developer Hackathon: ACT II Hackathon"

Team Idea

Deriving hardware-native GPU programming primitives from architecture graphs using ML — closing CUDA's 15-year co-design moat for AMD & others.

Team Members

Arihant Bansal

darkhacker

Himanshu Jain

himanshujain891356031

Submission

POLYFORGE

POLYFORGE is a universal parallel kernel compiler powered by LLMs and a zero-trust race-detecting oracle. Drop in a CUDA kernel, and POLYFORGE proves it correct and compiles it for custom parallel hardware like RISC-V GPUs without rewriting code.