Development of an Interactive Information Support Platform Integrating 3D City Models (PLATEAU) and Local LLM

Abstract

This study develops and validates an “Interactive Information Support Platform” that integrates the 3D city model “PLATEAU” with Large Language Models (LLMs) to resolve delays in evacuation behavior and administrative decision-making during disasters. The system synchronizes real-time shelter congestion data via QR codes (Firebase) and knowledge-graphed geospatial information (Neo4j/Fuseki) onto a digital twin base powered by Mapbox GL JS.A key technical feature is the implementation of a “Local LLM (Ollama)” to handle sensitive evacuation data securely and maintain resilience against network congestion, combined with “GraphRAG” technology to suppress hallucinations by anchoring responses in external knowledge. This allows the AI to function as an “Intelligent Operator”, capable of providing immediate, optimal evacuation routes and tailored advice by analyzing natural language inquiries within the context of dynamic congestion and geospatial constraints.This proposed system serves as a phase-free information infrastructure that bridges disaster response and daily urban management, realizing resilient decision-making support for smart cities.