X-Desert AI: An Agentic Framework for Detecting Areas Deficient in Opportunities and Services

Abstract

Analyzing access to essential community services and opportunities, such as healthcare, food, jobs, and education, requires substantial knowledge in spatial science and analytics. While R- or Python-based packages exist, in many cases they require sophisticated programming knowledge and offer less flexibility in scaling and reproducing analysis across different service types, study areas, and analytical parameters (spatial scale, travel time threshold, and travel mode). To streamline this analytical process, this study develops X-Desert AI, a multi-agent interactive web application that automates high-resolution service-desert (areas without access) analysis for any service type across the United States, at any given set of parameters. The app consists of a multi-tier agentic framework in the background where agents are led and coordinated by an LLM-based planner agent and operated by four agents tasked with data processing, network building, accessibility modeling, and visualization. In addition to managing and coordinating backend tasks, the planning agent operates on the frontend via a chatbot interface that serves as a dynamic assistant, guiding users through the process and reproducing the workflow as per user needs. As opposed to writing complex code, users can explain their analysis in plain English and visualize the output according to their preferences. The web application can support urban transportation, health, and community planning agencies by reducing programming burdens and enabling flexible, on-demand analyses.