ROCmPilot

ROCmPilot is an agentic developer tool for helping AI teams migrate CUDA/NVIDIA-first workloads toward AMD ROCm readiness. Users paste a public GitHub repository URL, and ROCmPilot runs a multi-agent audit across Dockerfiles, Python inference code, vLLM launch scripts, dependency files, health checks, and benchmark scripts. The system coordinates specialized agents: Repo Doctor finds CUDA and NVIDIA assumptions, Migration Planner creates ROCm-safe recommendations, Build Runner prepares validation steps, Benchmark Agent separates estimates from live AMD proof, Memory Agent stores reusable decisions, and Report Agent generates a judge-ready technical and business report. ROCmPilot currently uses Hugging Face and Qwen for model-backed report generation, with a fallback-safe demo path. It is designed to upgrade to AMD Developer Cloud by serving Qwen on AMD Instinct GPUs through ROCm and vLLM. The product is useful for infrastructure teams, AI startups, and open-source maintainers who want a faster, safer path to AMD GPU portability.