A novel approach to AI LLM Agents for interaction with Geospatial Databases
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
The integration of Large Language Models (LLMs) with geospatial databases is complicated by several practical limitations. First, LLMs operate on textual data and are not designed to perform precise computations over points, lines, and polygons. Second, the model’s context window is too small to accommodate query results that may contain tens of thousands of records or more. This paper proposes an architecture that addresses these challenges by separating responsibilities between system components. The LLM is responsible for interpreting the user’s query and planning the sequence of actions required to solve the task, while a specialized backend executes the geospatial operations. To store intermediate results, an external memory mechanism is employed: the model sends geospatial queries to the backend and receives only short descriptors (handles) in response. These descriptors serve as references to query results stored in external memory and can be reused in subsequent operations. This approach allows the LLM to construct complex chains of geospatial operations without exceeding its context limitations.
Fields of Study & Subjects
Data Management and Algorithms
Topic Modeling
Geographic Information Systems Studies
Publication details
- Journal / source
- COMPUTER-INTEGRATED TECHNOLOGIES EDUCATION SCIENCE PRODUCTION
- Issue
- 62
- Pages
- 99–106
- Publication date
- 28 March 2026
- Language
- EN