CHATBUS: Replicable Intelligent Chatbot Architecture for Information Quality in Urban Public Transportation
Public transport; Chatbot; LLM; Information Quality; Informational efficiency; RAG.
The fragmentation and low quality of information provided to public transport users undermine trust in the system, increase uncertainty regarding timetables, routes, and operational conditions, and reduce the informational efficiency of the service. In parallel, the advancement of Large Language Models (LLMs) and conversational architectures has expanded the possibilities for improving access to information through natural-language interactions. However, the literature lacks solutions capable of integrating official datasets, real-time operational information, and institutional content into a single accessible channel, as well as systematic evaluations of Information Quality (IQ) in LLM-based systems. In response to this gap, this research proposes and evaluates ChatBus, an intelligent chatbot designed as a computational artifact to centralize public transport information, initially applied to the bus system of São Paulo. The study adopts the Design Science Research (DSR) method and develops two architectural variants—single-agent and multi-agent—both integrating NLP, LLMs, Retrieval-Augmented Generation (RAG), and official APIs (SPTrans/Olho Vivo, Google Maps). The evaluation focuses on objective IQ metrics (accuracy, completeness, and timeliness) and informational efficiency indicators (response time and operational correctness rate). As a contribution, the work aims to deliver a replicable architecture and a reproducible experimental protocol, strengthening the use of AI to improve information systems in urban mobility.