LATInsights: LLM-Powered Conversational Agents for Geospatial Climate Analytics.

Urban climate adaptation planning requires synthesizing complex Earth Observation data, yet traditional Plan-ning Support Systems face an “implementation gap” due to technical barriers preventing non-specialist plan-ners from accessing specialized geospatial workflows. We present LATInsights, a conversational agent systemthat addresses this gap through natural language interaction, enabling municipal planners to perform sophisti-cated thermal analytics without remote sensing expertise. The system implements a novel two-tier architecture:a stateful LangGraph-based orchestration engine managing semantic reasoning and multi-turn dialogue, de-coupled from a stateless Model Context Protocol Server exposing domain-specific geospatial tools. This sep-aration enables independent scalability while ensuring analytical accountability through structured planningand three-tier error recovery with Large Language Model-driven re-planning. We demonstrate operationalefficacy through Milan’s Via Neera cycling corridor case study, where the system autonomously orchestratesmulti-scenario thermal simulations. This work demonstrates how conversational AI can democratize geospa-tial intelligence for evidence-based urban climate governance.