It works by functional decomposition as the first step. You break everything down into inputs, processes and outputs (IPO). Once everything is broken into these atomic units, the context is natively optimized for neural networks which need to be highly adaptive.