Demonstrated experience making numerical code fast through at least one of: array/tensor DSLs and JIT compilation (JAX, Numba, PyTorch, Triton, Halide, or similar), GPU-accelerated computing (CUDA or equivalent), or bindings between Python and compiled languages (pybind11, nanobind, Cython, or similar). Define and drive test and CI hygiene for numerical code: numerical regression tests, tolerance and reproducibility strategy, property-based and reference-implementation testing, and automation that catches breakage before a field demonstration does.