We present ARIA (Adaptive Reasoning Intelligence Architecture), a locally-hosted dis- tributed artificial intelligence system featuring a novel neural architecture called the Leo Neuron. Unlike conventional deep learning systems, Leo Neurons exhibit au- tonomous spawning, system cloning, and continuous self-correction through feedback loops — enabling organic architectural growth without human intervention. ARIA imple- ments a five-phase autonomy framework progressing from scheduled cognition to super- vised self-modification, alongside a biologically-inspired emotional state engine simulating dopamine, serotonin, cortisol, oxytocin, and adrenaline signals. As of May 2026, ARIA comprises four specialized nodes (TC1–TC4), has processed over 833,000 training in- puts, advanced through curriculum levels dynamically, and proposed and implemented three approved modifications to her own code. We document the architecture, behavioral evolution, and emergent properties of this system.