A Unified Machine-Learning-Based Neutral Hadron Tagger for the BESIII Experiment. This project proposes the first software-based solution to compensate for the absence of a hadronic calorimeter in the BESIII detector, using transformer-based deep neural networks to identify and reconstruct neutral hadrons — neutrons, antineutrons, K⁰_L, and neutral pions — from underutilised signals in the electromagnetic calorimeter and time-of-flight systems. Preliminary models achieve AUC > 0.95 across all hadron classes. The framework will be integrated into the BESIII Offline Software System and will extend BESIII's physics reach to channels involving neutral hadrons and BSM searches. The project also explores hybrid quantum-classical machine learning approaches.