A Well-Being-Centric Walkable Tourism Platform Using AI Agent

Abstract

We propose a Well-Being-Centric Walkable Tourism Platform Using AI Agent that fuses heterogeneous urban data—an OSM pedestrian graph, a city-scale microclimate field, and a walkability layer for Sofia, Bulgaria—with an AI agent that converts freetext user requests into a constrained heat-aware routing problem. The platform addresses a gap in existing digital map services, which offer rich navigation but lack integrated support for well-being-oriented urban exploration under intensifying urban heat islands. The routing engine implements two complementary strategies: P2, a ρ -detour-constrained heat-aware route obtained by Lagrangian relaxation, and PLLM, in which a large language model maps the user’s chat message and current temperature to a bounded detour budget ρ fed into the same engine. In a systematic evaluation over a POI grid in central Sofia, heat-aware routes consistently reduced cumulative heat exposure with only marginal detours, outperforming both the shortest-distance baseline and the heat-blind Google Directions API, while PLLM adaptively selected detour budgets matching user personas. We further conducted a subjective field study with local residents in Bulgaria, comparing the baseline shortest route with the proposed heat-aware route. The results indicated that the proposed routes were consistently feasible for walking and were perceived as more comfortable and scenically pleasant in several segments.