From Data to Dialogue: A Multi-Agent Chatbot Interface for Decision Support in Guide Dog Training

Guide dog training programs collect extensive behavioral, medical, and performance information for every dog, including the Behavioral Checklist (BCL), which can contain over 50 detailed assessment items per dog. Trainers rely on these records, together with data from thousands of past dogs, to judge whether a dog is on track, where it is struggling, and whether it is likely to succeed in the training program. Furthermore, although all this information is stored in organizational digital databases, trainers often can not use it in practice because the data requires technical processing, aggregation, and interpretation. To bridge these gaps, we present a multi-agent conversational chatbot that embeds the predictive model and the data processing pipeline inside a dialogue-based interface. Instead of navigating complex databases or analytical tools, trainers can ask natural-language questions such as “How does this dog compare to past dogs with similar profiles?” Our suggested chatbot retrieves relevant information, analyzes it using structured data and model outputs, and presents clear information that supports daily training decisions. We conducted a formative study with three professional experienced guide dog trainers across three main tasks: information analysis, dog comparison, and success prediction. Trainers reported that the chatbot made the underlying information more approachable, more interpretable, and easier to apply when making decisions about each dog’s training trajectory.