We describe a methodology for constructing testable theoretical physics frameworks using human geometric intuition as the primary input, artificial intelligence as the mathematical translator and computational verifier, and observational data as the selection mechanism. The method was developed through the construction of the Bulk-Caustic Engine (BCE) framework, which produced unfitted predictions matching observations across eleven independent physical domains in approximately one month of work by a single non-physicist collaborating with an AI system. Every equation, derivation, and numerical computation in the framework was produced by the AI; the human contributed exclusively geometric descriptions and directional guidance. We formalize the methodology as a three-component cycle — Shape, Translate, Test — and argue that it represents a distinct and potentially complementary approach to the traditional theoretical physics workflow. The key insight is that geometric intuition and mathematical formalization are separable skills, and their separation allows people with strong spatial/physical intuition but no formal mathematics training to contribute theoretical physics insights that can be rigorously tested.