GRX brings PyTorch-style tensor operations to Ruby. Every arithmetic op, activation, and optimizer step runs through a native C library with dynamic multi-target SIMD dispatch (AVX2+FMA, SSE, and scalar fallback). Ruby is the interface — C does the work. Features: autograd, SGD/Adam optimizers, Linear/Sequential/Dropout/BatchNorm layers, MSE/BCE/CrossEntropy loss functions, Xavier and He weight init. Cross-platform: .so on Linux, .dylib on macOS, .dll on Windows.