Research Statement

My research studies how large language models (LLMs) and large vision-language models (LVLMs) understand, reason about, and interact with geospatial information. Maps play a central role in how people understand places, environments, and spatial relationships. However, modern AI systems often struggle to interpret spatial relationships, incorporate geographic context, or answer questions that require multi-step spatial reasoning.

To bridge this gap, I design evaluation frameworks, benchmarks, and analytical methods that reveal not only where models succeed or fail, but why. I focus on advancing models’ capabilities in spatial reasoning, which includes understanding topological and directional relationships, interpreting map symbols and semantics, and integrating multimodal geographic context for more reliable decision-making.