OGD4All: A Framework for Accessible Interaction with Geospatial Open Government Data Based on Large Language Models

OGD4All is a Large Language Model framework that enables citizens to interact with geospatial open government data through natural language queries, achieving 98% analytical correctness and 94% recall while minimizing hallucinations. The system combines semantic retrieval, agentic reasoning, and sandboxed execution to provide transparent, auditable access to public datasets, representing a significant advance in making government data democratically accessible.

OGD4All addresses a critical gap in public data accessibility by leveraging LLMs to bridge the technical expertise barrier that typically prevents citizens from querying government datasets. The framework's architecture—combining semantic retrieval with sandboxed code execution—demonstrates a mature approach to deploying LLMs in high-stakes governance contexts where accuracy and auditability are non-negotiable. The 98% analytical correctness rate across diverse question types signals that LLMs can reliably handle factual queries while appropriately rejecting unanswerable questions, a capability that has historically plagued AI applications in regulated domains.

This work emerged from growing recognition that open government data remains inaccessible to non-technical populations despite its theoretical public availability. Traditional barriers—complex query languages, technical infrastructure requirements, and data fragmentation—have limited civic engagement with public information. By making 430 City-of-Zurich datasets queryable through natural language, OGD4All lowers friction for citizens, journalists, and researchers seeking transparency and accountability from government.

The framework's emphasis on explainability and reproducibility matters particularly for government technology. Because every query generates verifiable outputs through auditable code generation, users can understand precisely how conclusions were derived, addressing concerns about algorithmic opacity in public administration. This transparency builds institutional trust in AI-assisted governance decisions.

Looking ahead, similar frameworks will likely proliferate across jurisdictions seeking to enhance citizen engagement with public data. The technical validation demonstrated here—across multiple LLMs and diverse datasets—provides a template for organizations evaluating LLM deployment in governance contexts. Success here could accelerate adoption of AI infrastructure for democratic participation, though implementation will depend on overcoming institutional inertia and establishing appropriate governance structures around these systems.

• →OGD4All achieves 98% analytical correctness and 94% recall on government data queries, demonstrating LLM reliability for public administration use cases.
• →The framework's sandboxed execution and semantic retrieval architecture provides transparent, auditable access to public datasets while minimizing hallucination risks.
• →Natural language interfaces to government data lower barriers for citizens, journalists, and researchers to engage with transparency and accountability.
• →Successful validation across 11 LLMs and 430 datasets provides a replicable template for other jurisdictions implementing AI-assisted governance infrastructure.
• →Emphasis on explainability and reproducibility addresses critical trust concerns around algorithmic decision-making in public administration.