OGD4All: A Framework for Accessible Government Geospatial Data

OGD4All, a transparent, auditable, and reproducible framework based on Large Language Models (LLMs), has been presented. It is designed to enhance citizens' interaction with geospatial Open Government Data (OGD). The system combines several key features to provide more intuitive and reliable access to information.

Architecture and Functionality

OGD4All integrates semantic data retrieval, agentic reasoning for iterative code generation, and a secure sandboxed execution environment. This approach enables the production of verifiable multimodal outputs, ensuring the transparency and reliability of the results. The framework was evaluated using a benchmark of 199 questions, covering both factual questions and unanswerable questions, on a dataset consisting of 430 datasets from the city of Zurich. 11 different LLMs were used for the evaluation.

Performance and Reliability

The test results indicate that OGD4All achieves 98% analytical correctness and 94% recall. The system is able to reliably reject questions unsupported by available data, minimizing the risks of hallucinations. Statistical robustness tests and expert feedback confirm the reliability and social relevance of the framework. This approach demonstrates how LLMs can provide explainable and multimodal access to public data, promoting trustworthy AI for open governance.