Compiler that fuses and lowers ML operations to optimized kernels across CPU, GPU, and TPU backends, speeding model training and inference pipelines.