Multimodal Large Language Models for Smart City Analysis from Unstructured Data in Singapore

Abstract

Achieving sustainable urban development in high-density city-states such as Singapore requires planning frameworks that integrate heterogeneous data sources, capture interactions across multiple spatial and temporal scales, and respond to rapidly evolving environmental and socio-economic conditions. However, most urban modeling solutions treat isolated aspects independently, resulting in a fragmented perspective that severely limits their operational utility and adaptability in real-world scenarios. This study proposes a multimodal large language model (MLLM)-based framework tailored to sustainable urban development in Singapore, combining crucial urban resource inputs, outputs, and socio-economic factors. External influences arising from global political and economic dynamics are embedded to enhance the model’s practicality. The proposed architecture supports parallel multi-task learning, human-in-the-loop interaction, and real-time data updates, ensuring high precision and practical utility. Experimental results conducted using Singapore-specific datasets demonstrate the model’s outstanding performance in generating interpretable and actionable responses that remain consistent across diverse scenarios while advancing the Sustainable Development Goals (SDGs). This work serves as a stable and adaptive paradigm for applying multimodal artificial intelligence in urban governance and policy-making.