Generative AI solutions company Xenon (CEO Seok-Tae Ko) announced on the 11th that it has introduced a generative AI-based conversational interface to Seoul's 3D spatial information platform, "S-Map." S-Map is a platform that integrates 3D spatial data across Seoul and serves as digital infrastructure for urban analysis and administrative decision-making. S-Map, featuring generative AI-based conversational capabilities, has been operating as a pilot service since the 23rd of last month.
This project was conducted as part of the "AI-Based Conversational Interface Application Digital Twin Service Demonstration Project," launched in 2024. The goal was to improve usability by converting S-Map's existing menu-centric structure to a natural language-based interface. Xenon served as the lead research and development organization, with spatial information company Gaia3D participating as a joint research and development organization.
Xenon has transformed the existing complex menu structure into a natural language-based conversational environment, making the platform more accessible to users of all ages. When users enter commands via voice or text, AI analyzes their intent and automatically executes actions such as moving the map, changing the viewpoint, searching for addresses, and controlling 3D buildings, road facilities, and lifestyle information layers. Furthermore, considering accessibility for first-time users, a chatbot-style user guide is also provided.
This system goes beyond simple information retrieval and is designed to enable advanced functionality using natural language commands. Users can perform functions such as comparing urban time series through split-screen, analyzing urban environments such as landscapes and wind paths, simulating real estate transaction prices, and supporting building design based on regulations, all without navigating menus.
To achieve this, Xenon applied its own proprietary prompt engineering (ReAct·DSPy)-based feature mapping AI model. This model analyzes users' natural language requests and connects them to various S-Map functions. Furthermore, by establishing an LLMOps system to manage generative AI operations, Xenon ensured both service stability and performance.
Performance evaluations also showed that the system met its target criteria. All evaluation indicators established in user evaluations and official certification tests were met, and the generative AI-based feature matching accuracy reached 81%, exceeding the target (over 70%). The average time from voice command to feature matching was 1.9 seconds, faster than the target of less than 3 seconds. The voice command recognition error rate was 1.7%, and user satisfaction reached 81%.
This demonstration project is also considered significant in terms of improving digital accessibility. Even users unfamiliar with using a mouse or keyboard can utilize digital twin services using voice or text input, lowering barriers to information access. Furthermore, natural language-based data exploration is expected to contribute to the intuitive provision of complex spatial information, improving the quality of citizen services and the efficiency of administrative decision-making.
“This S-Map demonstration project is an example of generative AI expanding to the stage where it actually controls public system functions,” said Ko Seok-tae, CEO of Xenon. “By meeting all evaluation indicators in the official certification test, we have confirmed the possibility of stable operation in a public environment.”
Meanwhile, in the public sector, smart city projects are expanding, combining digital twins and generative AI technologies to improve the efficiency of urban management and administrative services.
- See more related articles
You must be logged in to post a comment.