In California, highway bridges are mainly multi-span box girder bridges. These types of bridges generally have high torsional rigidity and bending capacity, low self-weight, and cost efficiency. California is located in the Pacific Ring of Fire, which is an area characterized by a high level of seismic activity due to tectonic plate boundaries. This region experiences frequent earthquakes because it sits along the boundary of the Pacific Plate and the North American Plate. Therefore, seismic design and post-earthquake assessment of bridge structures are particularly important. Many highway bridges in California are located in remote areas, and after an earthquake, engineers often take high-definition images of bridge details, such as bearings and piers, to see if the bridge's load-bearing capacity meets standards. By integrating diverse datasets, including seismic ground motion data, historical earthquake records, and real-time structural health monitoring data, the proposed generative artificial intelligence model aims to simulate various earthquake scenarios and their impacts on bridge structures. This research utilizes advanced machine learning techniques, including Generative Adversarial Networks and Variational Autoencoders, Large language Models, to generate predictive models that provide insights into damage mechanisms and facilitate informed decision-making for maintenance and repair strategies. This thesis has developed a generative artificial intelligence model to evaluate the seismic performance and post-earthquake load-bearing capacity of multi-span box girder bridges. This model features text recognition and image recognition capabilities. By uploading images of damaged bridge components or data texts (such as seismic data, bridge inspection reports, structural health monitoring data), the model can compute and derive the corresponding results.