Agentic AI Turns Transit Data Into More Targeted Decisions

In Salvador, Brazil, researchers developed an agentic artificial intelligence system called SUNTInsight to provide transit managers with targeted operational decisions from a complex transportation dataset. The dataset included information on approximately 700,000 passengers, 2,000 vehicles, 400 lines, and 3,000 stops and stations.

Traditional methods of analyzing such large amounts of data often relied on static dashboards and manual queries, creating a gap between the data collected and the insights managers could use. SUNTInsight aimed to bridge this gap by allowing managers to explore the data using natural language. The system was designed to translate these requests into structured query language (SQL), retrieve the relevant information, and display the results along with visualizations.

This conversational interface enabled managers to understand what the AI system was doing and make more informed decisions. During testing, researchers found that broad averages could conceal problems occurring in specific parts of the transportation network. A request for data on a particular bus route during a morning rush period revealed that occupancy varied along the route, with heavier demand concentrated in certain neighborhoods.

This finding led to more targeted operational measures, such as adding buses in high-demand areas and using short turns on congested sections. SUNTInsight faced challenges during development, including issues with tool calling and security concerns. However, larger reasoning-focused models performed more consistently. The system ultimately provided a practical co-pilot for transit management, illustrating how agentic AI could connect operational data with targeted decision-making while maintaining human oversight.