GuardianMind: A Multi-Modal Enhanced Large Language Model for Smart City Emergency Response
The escalating complexity of urban emergencies, driven by rapid urbanization and climate change, highlights the critical need for advanced emergency response systems. Traditional methods, reliant on manual judgment and fragmented information, struggle to meet demands for rapid, precise, and efficient incident management. While large language models (LLMs) offer potential, general-purpose LLMs exhibit limitations in information timeliness, domain expertise, multi-modal data integration, and decision support accuracy within smart city emergency response. To address these challenges, we propose GuardianMind, a novel multi-modal enhanced LLM system specifically engineered for smart city emergency response. GuardianMind integrates a powerful base LLM with specialized modules: a City Emergency Knowledge Retrieval component, a Smart City Knowledge Graph, a Real-time Data and Tools module, and a Public Information Search module. This architecture enables GuardianMind to effectively process and synthesize diverse, heterogeneous data streams, providing a holistic understanding of emergencies and generating professional, accurate, and actionable response suggestions. Through comprehensive experiments on a custom-built dataset, GuardianMind consistently outperforms state-of-the-art general LLMs, including leading commercial and open-source models, across critical dimensions of accuracy, professional depth, and timeliness, while maintaining excellent language fluency. An ablation study further validates the indispensable contribution of each integrated module. Our qualitative analysis demonstrates GuardianMind’s capacity to deliver highly precise, context-rich, and immediately actionable intelligence, marking a significant advancement in intelligent urban crisis management.
