The Hidden Layer

KernelPilot is a VS Code extension that helps developers explore and evaluate GPU optimization opportunities in C/C++ and CUDA code. It combines Clang-based source analysis with Meta’s Muse Spark 1.3 for code explanations and model-assisted optimization suggestions. Developers can review proposed CUDA candidates, compile and validate them against a reference, and compare CPU and GPU performance using workload-specific benchmarks. KernelPilot treats AI recommendations as hypotheses rather than verified outcomes: correctness checks and measured results provide the evidence for evaluating an optimization.