GeoNAV-LLM: Semantic Modelling and Geographic Question Answering
Designed a Geographic Large Language Model (GeoNAV-LLM) to translate natural-language geographic questions into structured spatial representations and analytical intents.
Developed a question decomposition framework that identifies spatial entities, relationships, constraints, and required geo-analytical operations from expert-formulated queries.
Integrated symbolic representations (ontologies, spatial concepts) with LLM-based reasoning to bridge linguistic expressions and geospatial data models.
Implemented semantic linking between maps, satellite imagery, vector layers, and spatial databases to support context-aware geographic reasoning.
Explored hybrid AI approaches, combining data-driven language models with explicit geographic semantics to improve interpretability and robustness.
Addressed limitations of purely statistical GeoAI models by incorporating explicit representations of geographic intent and workflow logic.
Demonstrated applications in urban and environmental analysis, enabling users to navigate from high-level questions to executable geo-analytical workflows.
Contributed toward the broader goal of formalising geographic questions as first-class computational objects, supporting explainable and reproducible spatial analysis.