Urban air quality research increasingly relies on integrating heterogeneous data from sources such as monitoring stations, satellite imagery, community sensors, built-environment datasets, and meteorological forecasts. However, researchers often face challenges in discovering, understanding, and integrating these datasets due to inconsistent metadata, mismatched spatial and temporal resolutions, diverse formats, and varying semantics. Moreover, effective use of air quality analytical models typically requires specialized expertise to interpret model requirements and identify suitable input datasets, leading to labor-intensive, inefficient workflows. This paper introduces LLM-AQRA, a prototype Large Language Model (LLM)-driven Air Quality Research Assistant designed to streamline these processes by automatically generating standard-compliant metadata, extracting model requirements, assessing dataset-model compatibility, and supporting map-based exploration of heterogeneous datasets. Initial user feedback indicates that LLM-AQRA is perceived as useful for dataset understanding and has the potential to reduce manual effort in analytical workflows. Future work will focus on retrieval augmented generation, ontology-guided constraints, expanded domain support, and broader user evaluations to improve reliability and applicability for real-world environmental analysis.