shashank hirwani@shashankh366207

Enterprise AI workloads face massive bottlenecks due to a heavy reliance on legacy NVIDIA CUDA codebases. Manually migrating these massive codebases to AMD ROCm takes months of specialized engineering time. Basic portability tools often lack the architectural awareness to optimize code for AMD Instinct architectures, leaving teams without visibility into actual hardware compilation and execution. RadeonShift AI solves this CUDA lock-in by providing an AI-assisted kernel migration assistant that translates CUDA to HIP/ROCm while auditing architecture-specific migration risks. It leverages a Syntax Translation engine hosted on Fireworks AI and a Mixture-of-Agents (MoA) Audit powered by DeepSeek-V4 to scan for PTX risks, hardcoded warp sizes, and AMD-specific optimizations. A Deterministic Rules Engine guarantees correctness by explicitly flagging unsupported features (like specific Tensor Core or async-copy operations) for manual redesign rather than hallucinating unsafe code. Additionally, when the optional hardware backend is online, the platform connects to a remote bare-metal AMD MI300X environment to surface live ROCm telemetry, compile-check evidence, and runtime benchmark provenance. Built with a Next.js frontend and a FastAPI hardware engine, RadeonShift AI shrinks initial audit and translation times from weeks to minutes, delivering a massive ROI for migration triage.

To participate in the hackathon, simply click Sign up with AMD. If you are not already a member of the AMD AI Developer Program (ADP), you will need to create an account. ADP members can access AI tutorials, experts, and community support. In addition, new members can claim a free month of DeepLearning.ai pro, $100 in AMD GPU cloud credits, and $50 in Fireworks AI API credits. All hackathon participants will receive on hackathon launch additional compute and API credits for their projects. Ready to build what's next on AMD? Sign up with AMD below to secure your spot.