Ensuring the seismic resilience of concrete structures in public buildings is a critical challenge in modern structural engineering, requiring innovative and interdisciplinary solutions. This study introduces a comprehensive framework that integrates Large Language Models (LLMs) and deep learning techniques to optimize seismic performance. By leveraging LLMs to analyze vast datasets of structural, material, and seismic information, and employing hybrid deep learning models for precise seismic response predictions, this approach bridges structural engineering, machine learning, and data science. The framework also incorporates reinforcement learning to dynamically optimize structural designs, enabling adaptive strategies for improving resilience under diverse seismic scenarios. Experimental validation on multiple datasets demonstrates significant improvements in prediction accuracy, structural performance, and computational efficiency compared to traditional methods. This research highlights the transformative potential of combining advanced computational intelligence with domain-specific expertise, paving the way for innovative applications in seismic engineering and public infrastructure design.