In this paper, we present a cutting-edge approach that combines Graph Neural Networks (GNNs) with AutoML for reconstructing ground-based cosmic ray (CR) observational data. Our novel method accurately estimates primary cosmic ray energy and enhances P/gamma identification. Leveraging Full Monte Carlo simulations, emulating the Tibet ASgamma experiment (Tibet-III + MD), we achieve compelling results. By harnessing the power of AutoML and GNNs, our integrated approach achieves a remarkable 31% enhancement in energy resolution for reconstructed cases above 100 TeV, surpassing the performance of S50 reconstruction. Additionally, our method effectively reduces the cosmic ray background by 30%, while preserving the crucial gamma events. The outstanding accuracy of our GNN-based energy reconstruction is further amplified through AutoML, which enables the assimilation of critical information, such as air shower size, secondary cosmic ray lateral distributions, density distributions on the detector, core position, zenith angle distributions, and more. Beyond cosmic ray observation, our versatile machine learning approach holds promise for tackling a wide range of particle physics and astrophysics challenges, making substantial contributions to these fields and paving the way for exciting future advancements.