Developer builds LLM-free AI brain in NumPy with spiking neurons and homeostasis

A software developer has published a detailed technical follow-up describing the internal architecture of a custom cognitive system built entirely in NumPy, without any external large language model. The system simulates biological brain regions including a brainstem, thalamus, hippocampus, and prefrontal cortex, each implemented as discrete modules with real code. A key design principle is an inverted-U arousal function inspired by the Yerkes-Dodson curve, where both low and high arousal reduce cognitive performance. Each processing pass moves through ten sequential stages, from arousal integration to memory storage, with outputs from one module feeding directly into the next. The project aims to demonstrate that memory, drive, and contextual reasoning can be achieved using owned weights and explicit neural architecture rather than rented API-based models.