What does Luminal do?
Luminal builds a compiler and runtime for AI inference that treats compilation as a search (superoptimization) problem, automatically compiling models into native code optimized for GPUs, ASICs, and other hardware by auto-generating and searching high-performance kernels. It aims to push GPU utilization well above the typical 10-20%, delivering 2-3x faster inference while preserving the standard PyTorch developer workflow. A Y Combinator (S2025) company, it raised a $5.3M seed in November 2025 led by Felicis Ventures.