Abstract

Large language models (LLMs) are a transformational capability at thefrontier of artificial intelligence and machine learning that can supportdecision-makers in addressing pressing societal challenges such as extremenatural hazard events. As generalized models, LLMs often struggle to providecontext-specific information, particularly in areas requiring specializedknowledge. In this work, we propose a Retrieval-Augmented Generation(RAG)-based multi-agent LLM system to support analysis and decision-making inthe context of natural hazards and extreme weather events. As a proof ofconcept, we present WildfireGPT, a specialized system focused on wildfirescenarios. The architecture employs a user-centered, multi-agent design todeliver tailored risk insights across diverse stakeholder groups. Byintegrating domain-specific projection data, observational datasets, andscientific literature through a RAG framework, the system ensures both accuracyand contextual relevance of the information it provides. Evaluation across tenexpert-led case studies demonstrates that WildfireGPT significantly outperformsexisting LLM-based solutions for decision support in natural hazard and extremeweather contexts.