AI-Powered Analysis

Agentic LLM Framework for Generating Spatial Intelligence to Support Decision-Making in Smart Cities

Smart cities generate multi-source, multi-timescale datasets through traffic sensors, transit operations, land use registries, sociodemographic surveys, crash reporting systems, and environmental monitoring. These datasets hold significant potential to help decision-makers make better choices about transportation systems, yet their heterogeneity and complexity hinder their potential from being fully harnessed. Existing data visualization platforms have advanced exploratory analysis capabilities, but are often rigid, designed for single datasets or fixed workflows, and provide limited support for scenario-based planning across different contemporary transportation problems. Recently, Large Language Models (LLMs) have been explored in smart city applications; however, they are not well-suited to handling structured spatiotemporal data and often lack interpretability in outputs. To address these challenges, this study proposes an agentic LLM framework that formalizes visualization as an intermediate reasoning layer between heterogeneous datasets and planner-facing decision support to generate spatial intelligence. The framework comprises four coordinated agents (supervisor agent, visualization agent, analysis agent, and metrics agent) that replicate expert planning workflows and generate interpretable, evidence-backed insights. It is validated for the City of Peachtree Corners, GA, a smart city testbed where ten datasets spanning transit ridership, traffic conditions, land use, and community facilities were integrated to generate spatial intelligence for decision-making. Two case studies demonstrate the framework's capabilities; the first evaluates how expanding the reach of the existing autonomous vehicle shuttle system can improve accessibility, and the second illustrates how interactive large mobile displays can be placed to enhance information access for residents. Together, these results highlight how the framework lowers technical, analytical, and interpretive barriers for decision-makers, while producing actionable recommendations.