"KernelOptix team on AMD AI League Hackathon"
Team Idea
Building an autonomous AI agent that analyzes, ports, and auto-optimizes custom GPU kernels (PyTorch/Triton/C++) specifically for AMD ROCm architectures. Using an LLM-guided profiling and compilation loop, the tool translates unoptimized compute primitives into high-performance HIP/ROCm kernels, benchmarks execution latency on AMD hardware, and iteratively refines code for peak compute throughput and lower memory footprint. Designed to streamline AI operator deployment across AMD Instinct and Radeon GPUs.
Team Members
.png&w=256&q=75)
Unknown Hacker
khatranfx18298
Submission
Submission draft in progress