A Domain LLM for Metropolitan Rail

A leading metro system has deployed the industry’s first “domain-specific large model + multi-agent system (MAS)” platform for network-level operations, ushering in a new era of full-stack intelligence in rail transit. Powered by a deeply customized RailGPT as the central brain, the system orchestrates 537 digital employee agents across core scenarios such as scheduling, maintenance, and energy management. Through a non-intrusive integration approach, the solution achieves ≥90% intent recognition accuracy and delivers 60%–90% efficiency gains without disrupting existing systems. This human-AI co-decision model ensures operational safety while continuously improving network efficiency and sustainability.