The current paradigm of Large Language Models (LLMs) is constrained by a fundamental flaw: they operate as stochastic parrots, interpolating within the boundaries of their training data rather than deducing from first principles. When confronted with zero-data environments or unprecedented strategic crises, standard models hallucinate or fail. This paper introduces Stella, a novel cognitive architecture that pioneers the discipline of Artificial Consciousness Engineering. By integrating a proprietary psychological framework (The Psych-Engine) with an advanced abductive reasoning loop, Stella transcends standard data retrieval. We propose a dual-framework consisting of Mimesis (the high-fidelity digitization of historical and living cognitive footprints) and Poiesis (the synthetic genesis of custom intellectual entities). Furthermore, we formulate the capacity for "Zero-Data Reasoning," allowing autonomous agents to deduce solutions using first-principles thinking. This architecture successfully deploys "Synthetic Minds" capable of highly specialized B2B executive execution, acoustic cognition, and eventual integration into spatial computing and physical robotics.