Researchers at McGill University propose a three-agent workflow for travel behavior modeling and weather-sensitive demand prediction. The approach integrates conversational data collection, structured data processing, and behavioral prediction using a chatbot-administered survey and machine learning models. The study evaluates the performance of nine large language models under different prompt-and-context conditions and finds that visual context can provide additional predictive information for selected models. --- Why it matters: This research matters to engineers working on AI-powered transportation systems because it presents a novel approach to integrating conversational data collection, structured data processing, and behavioral prediction. The study's findings on the effectiveness of multimodal large language models in predicting travel behavior under different weather conditions can inform the development of more accurate and robust transportation systems. Source: https://arxiv.org/abs/2608.20320