Understanding travel behaviour is crucial for designing adaptive and inclusive urban transport systems, especially in ageing societies. Older adults often face unique mobility barriers that traditional survey models fail to capture. Classical machine learning methods, while effective for structured prediction, struggle to include contextual nuances and unstructured feedback.

This thesis explores a new approach to travel behaviour analysis using a modular, multi-agent system powered by large language models. The proposed framework combines several specialised agents, each handling a specific sub-task such as summarisation, classification, or reporting, and coordinates them through a shared data context. The pipeline was applied to real-world data from Malmö’s elderly population, producing structured reports that integrate personal, infrastructural, and environmental contextual factors.

Outputs were evaluated through a blinded expert review involving specialists in urban planning and data science. Results showed that the pipeline generated policy-relevant and semantically coherent insights comparable to those written by human experts. Key strengths include modularity, low labour requirements, and transparency.

Limitations include the lack of real-time data integration and the reliance on commercial APIs, but the study offers a practical proof of concept. By combining human oversight with emerging AI capabilities, this work demonstrates a new direction for urban analytics, where people and machines can jointly produce timely, explainable, and context-aware transport planning tools.