Static-allocation MLP inference in ANSI C using a 2-slot ring buffer

This project provides a minimal, static-allocation implementation of a multi-layer perceptron (MLP) inference engine in ANSI C, designed for microcontrollers with severe RAM constraints. By using a 2-slot ring buffer to reuse memory across layers, it eliminates dynamic allocation entirely, making it ideal for embedded developers who need deterministic, low-memory neural network inference. It is interesting because it demonstrates a practical, years-long exploration of extreme memory optimization for tiny AI, achieving inference with only two activation buffers regardless of network depth.