The emergence of the AI for Science research paradigm has endowed artificial intelligence (AI) with the ability to drive scientific discovery and the potential to solve complex engineering problems. As one of the largest producers of scientific data, synchrotron radiation sources provide a vast stage for AI applications. The beamline software system of the High Energy Photon Source (HEPS) has pioneered a new research paradigm that combines "AI for Science" with a "Unified Software Framework Bringing Facilities Together". This system systematically conducts research on the foundation algorithms of AI large models for specific methodologies. By embedding interpretable physical information, it enables the training and deployment of physical discipline large models. Coupled with the unified data acquisition software system and ecosystem of the light source, it is expected to become a paradigm of "AI for Science" research driven by algorithm and software optimization.