I need a streamlined Alpine Linux build with a custom-compiled kernel tuned for raw performance on AI workloads. The image must stay lightweight yet still include hardened kernel options and built-in data-at-rest encryption so I can safely move it between research environments. My toolchain will centre on TensorFlow, PyTorch and scikit-learn, so whatever patches, drivers or extra modules you add have to play nicely with CUDA/cuDNN and common BLAS back-ends. I am happy with either a reproducible Dockerfile or a bootable ISO—whichever gives the cleanest path to deployment on bare metal and in KVM. Deliverables • Alpine Linux image (ISO or container) with custom kernel, GRSEC-style hardening flags enabled and default LUKS support • Kernel config / build scripts committed to Git so I can rebuild on demand • Minimal package set plus working, documented install steps for TensorFlow, PyTorch and scikit-learn on the final system • Short README that explains every optimisation, security tweak and how to extend the image in future I’m available to clarify hardware targets or additional drivers the moment you start. Let’s build something fast, secure and reproducible.