The design of woven composite materials is inherently complex and typically requires specialized expertise in micromechanics, geometric modeling, and computationally intensive finite element analysis. Conventional design workflows often rely on repeated high-fidelity simulations and trial-and-error strategies, resulting in high computational cost, limited interactivity, and restricted accessibility for non-expert users. This work presents an integrated artificial intelligence framework that combines large language models with surrogate-assisted reinforcement learning to enable intelligent and efficient woven composite design. In the proposed system, a large language model acts as a high-level reasoning and interaction agent that interprets unstructured natural language inputs and translates user intent into formal design objectives, constraints, and optimization parameters. A Deep Q-Network–based reinforcement learning agent then explores the high-dimensional design space, optimizing both discrete variables, such as material selection, and continuous parameters governing weave geometry and architectural configuration to improve mechanical performance. To mitigate the computational burden of iterative finite element simulations, a physics-informed surrogate model is incorporated to rapidly predict elastic–plastic material responses, including full stress–strain behavior. The framework also integrates a parametric three-dimensional microstructure generation module that enables real-time visualization of optimized woven architectures while accounting for physically relevant effects such as yarn flattening, shear deformation, and realistic inter-yarn interactions. Numerical results demonstrate that the proposed approach reduces design iteration time from hours to seconds while maintaining predictive accuracy and geometric fidelity. By coupling natural language interaction with physics-aware optimization, this framework streamlines the woven composite design process and lowers the barrier to advanced composite engineering for academic and industrial applications.