Fine-Tuning Generative AI for Geospatial Healthcare Solutions through Python Query Execution

Keywords: GeoAI, ChatGPT, Spatial Query, Python, LLM, Generative AI
Abstract Type: Paper Abstract

Abstract

Generative AI and LLM models like ChatGPT have advanced rapidly across diverse fields but continue to struggle with complex geospatial tasks, especially those demanding spatial precision and problem-solving. This study examines the limitations and possibilities of using generative AI to address geospatial challenges by fine-tuning a ChatGPT model on a specialized healthcare dataset to solve python queries. Our approach enables the fine-tuned model to generate Python code for answering intricate spatial questions—such as buffers, spatial joins, and driving times—which is executed immediately to provide real-time solutions in a conversational way. By iteratively refining the model on healthcare-specific geospatial queries, we enhance its ability to offer actionable insights for healthcare planning and accessibility assessment. Initial results indicate significant improvements in the model's spatial reasoning, showing that this code-generating, fine-tuning approach on domain-specific data can bridge performance gaps in generative AI for geospatial applications. This research demonstrates a promising new approach for integrating GeoAI into healthcare accessibility and geospatial analysis. By refining generative AI models to tackle spatial queries, via python, with accuracy and immediacy, we open the door to AI-driven tools that empower data-informed decision-making and improve resource allocation in healthcare. Our findings suggest a future where AI not only supports but actively advances spatial problem-solving in complex, real-world applications.