This project explores the use of Large Language Models (LLMs) to support geospatial analysis by integrating them with data processing and decision-making pipelines. We focus on task decomposition, spatial reasoning, and workflow automation to enable natural language interaction with GIS operations. The goal is to evaluate how LLMs can assist in formulating, executing, and interpreting, and executing geospatial queries with the ultimate objective of making complex analyses more accessible and reproducible. In fact, our research involves large-scale experimentation with pre-trained and fine-tuned LLMs, retrieval-augmented generation from structured and unstructured geospatial metadata, and benchmarking model capabilities across tasks such as spatial querying, data harmonization, and geospatial task completion. Due to the computational demands of training, fine-tuning, and evaluating state-of-the-art LLMs and and multi-agent frameworks on geospatial tasks, access to high-performance computing resources at NAISS is critical. The results are expected to contribute to both methodological advances in GeoAI and practical tools for more transparent, accessible, and efficient geospatial data use across research, education, and policy-making.