Broadening Access to Transportation Safety Data with Generative AI: A Schema-Grounded Framework for Spatial Natural Language Queries

AI can democratize access to safety data without sacrificing accuracy: The framework uses generative AI to translate plain-language questions into structured database queries, allowing non-technical users like school committees and residents to analyze transportation safety data that was previously accessible only to GIS specialists.

Separating language interpretation from execution preserves reliability: By having the LLM generate semantic frames that are then validated through rule-based checks before querying a PostGIS database, the system maintains reproducible, auditable results—addressing key concerns about using generative AI in public-sector decision-making where transparency matters.

Schema-grounding prevents AI hallucinations in critical safety contexts: The framework constrains the AI to work within defined data structures and validated rules rather than allowing free-form outputs, reducing the risk of the model generating plausible-sounding but incorrect safety analyses that could influence real-world planning decisions.

Technical barriers are a significant equity issue in safety planning: The paper identifies that local agencies, school committees, and community residents often have legitimate safety concerns but lack the technical capacity to access relevant crash records and geospatial data, suggesting that removing these barriers could enable more inclusive participation in transportation safety decisions.