Date of this Version

11-19-2025

Document Type

Presentation

Abstract

Geospatial research workflows involve heterogeneous data, specialized platforms, and multi-step processes that traditionally depend on extensive human expertise. This slide deck presents a practical exploration of how agentic AI can support such environments—not by replacing established workflows, but by augmenting them with structured, reliable assistance. Using FIU’s EnviStor project as a case study, we examine how a multi-agent system—built on a 3-track engineering knowledge architecture of behaviors, domain concepts, and procedural skills—can collaborate with human analysts to handle data preparation, metadata work, and platform operations across systems such as Dataverse, ArcGIS, and Pelican.

Rather than aiming for universal conclusions, this work highlights the lessons learned from deploying agentic AI in a controlled, real-world GIS context. The focus is on understanding what makes an AI agent reliable—including consistency, memory, safety protocols, and interpretable workflows—and how structured knowledge can guide an agent’s reasoning. Through these early tests, we observe how agents and humans can meaningfully share work, how agents learn from real tasks, and where current limitations remain, particularly around validation, error recovery, and workflow robustness.

This project represents a first step toward exploring agentic AI in geospatial research. The goal is not to generalize beyond the project, but to surface practical insights, identify emerging patterns, and outline the challenges that must be addressed before such systems can be considered dependable components of scientific data management.

Download

Share

COinS