Spatially-Aware Conversational AI for Real Estate: Enhancing Query Understanding Using Location-Aware Language Models

Abstract

This paper introduces a spatially-aware conversational AI system designed for real estate platforms, which uniquely combines natural language processing (NLP), geospatial intelligence, and user behavior to improve query understanding and property relevance. Traditional chatbots often struggle to interpret vague or spatially contained queries such as “apartments near Nugegoda under 40 million,” leading to mediocre user experiences. Our solution addresses this limitation by integrating fine-tuned locationaware large language models (LLMs) with IP-derived geolocation, GIS proximity data, and contextual query interpretation. The system uses both structured and unstructured user inputs to clarify preferences, infer missing details, and deliver contextually filtered property results. It also supports live dialogue for clarifications, enhancing interactivity and accuracy. We detail the architecture, training process, and spatial NLP techniques used to ground language understanding in real-world locations. Internal evaluations on a Sri Lankan property dataset show marked improvements in intent classification, query
resolution accuracy, and user satisfaction. The findings demonstrate that embedding spatial reasoning within AI chat systems can significantly elevate real estate platform usability, opening new possibilities for localized and intelligent property discovery