The Centre of Innovation for Data Tech and Artificial Intelligence (CIDAI) invites you to the results presentation session of the project Intelligent Mobility Agent for ATM, an initiative aimed at exploring the potential of generative artificial intelligence applied to public mobility services in Catalonia.
The project has developed an intelligent conversational assistant capable of answering questions in natural language about public transport in Catalonia, integrating both static data and real-time operational information. The solution leverages the open standards GTFS and GTFS-RT to provide contextualized, reliable, and accessible information to citizens, public administrations, and operators.
The Intelligent Mobility Agent represents a new approach to semantic access to public mobility data, combining generative artificial intelligence technologies with traditional data querying and exploitation architectures. The system incorporates a language model integrated through the OpenAI API and implements the ReAct pattern (Reasoning + Acting), which enables the combination of reasoning capabilities with the execution of specific actions on tools and databases.
The developed architecture follows a modular and scalable approach, integrating a conversational interface, a FastAPI-based backend, a database optimized for large volumes of information, an interactive dashboard, and multimodal features such as Speech-to-Text through Whisper. The system allows users to query schedules, routes, connections, incidents, nearby stops, or operational alerts using natural language.
During the session, the main technical and functional results of the project will be presented, along with the challenges identified related to data quality, response latency, and service scalability.
From ATM’s perspective—particularly from the Mobility Information Management Center (CGIM)—this project represents a first structured approach to the use of generative artificial intelligence in the mobility domain, helping to identify both its potential and the technical and functional constraints of its application in real operational environments.
