Exploring machine-learning primitives implemented purely in C for maximum performance and portability.
ML in C is an exploration of what machine-learning primitives look like when you remove every abstraction layer. No Python, no framework, no automatic differentiation engine — just C, a compiler, and the math. The goal is to build a working understanding of the fundamental operations that underpin neural networks by implementing them directly.
Most ML education happens through frameworks that abstract away the interesting parts. Knowing that model.backward() computes gradients is not the same as knowing how it actually works. Implementing backpropagation, matrix operations, and activation functions in plain C forces a complete understanding of the mechanics — and reveals exactly where the computational cost comes from.
There's also a practical angle: C ML runs anywhere. No Python environment, no CUDA dependency, no package manager. A model compiled with this library should run on whatever hardware is available.
Early exploration. Currently investigating matrix operations, activation functions, and the structure of a forward pass. A write-up will follow once the fundamentals are solid enough to be worth documenting.