We propose an autonomous protein-design system inspired by the REvolve framework (ICLR 2025), which uses Large Language Models (LLMs) as evolutionary operators to generate and mutate reward functions, with human feedback providing the fitness signal for selection. In our setting, we transfer the same core idea—LLM-driven evolutionary search over a structured design space—but operate on protein binder sequences rather than reward functions and use physics-based simulations rather than human feedback as the fitness signal. The Epidermal Growth Factor Receptor (EGFR) extracellular domain is fixed as the target (Chain A), and in each generation, the LLM proposes candidate binder sequences (Chain B). AlphaFold-Multimer predicts the 3D structure of the EGFR–binder complex, docking refines the binding pose, and short GPU-accelerated molecular dynamics simulations quantify folding stability and binding persistence. These simulation-derived metrics are combined into a scalar fitness value that drives a REvolve-style evolutionary loop: high-fitness binders are selected, used to condition the next LLM calls, and mutated into new candidates. This closed-loop pipeline—LLM generation, structure prediction, docking, simulation, scoring, and evolutionary selection—aims to enable scalable, fully automated discovery of EGFR-targeting protein binders without human expert intervention in the inner loop.