Agentic AI for Crisis Informatics: A Multi-Agent Framework for Scalable and Reliable Disaster Communication

The increasing frequency and severity of disasters have increasingly overwhelmed traditional response capabilities, necessitating a shift from passive tools to autonomous agentic systems. Whilst current AI excels at specialized tasks, a research gap remains in integrating these capabilities into a cohesive information system (IS) capable of independent reasoning and action, designed for intelligent human-AI collaboration. This paper proposes a novel conceptual architecture for a dedicated Environmental Question-Answering System (ENV-QAS), a hierarchical Agentic AI framework unifying a Multi-modal Large Language Model (MLLM), Temporal Knowledge Graphs (TKG), and Retrieval-Augmented Generation (RAG). Grounding our framework in the Uses and Gratifications Theory (UGT), ENV-QAS aims to enable real-time disaster decision-making and enhance human-AI collaboration in the field through accessible decision support. By bridging data gaps and enabling multi-agent collaboration within a centralized reasoning engine, this study contributes to the next generation of IS artifacts. As a design-science artifact, ENV-QAS is accompanied by a concrete evaluation plan positioning it for empirical validation in future work.