Scheduling Efficiency Up 80%: SenseTime's Large Device Helps Launch Shanghai Quantum City's First AI Agents

SenseTime's Large Device and the Shanghai Municipal Bureau of Planning and Natural Resources upgraded the co-developed 'Yunyu Starry Sky Large Model', moving it from answering questions to executing tasks. The bureau unveiled the first government agents of the Quantum City, including the flood-control agent 'Yuzhi Heluo', the livelihood service agent 'Meihao Shenghuo Zhangzhongbao', and the Jiangnan tour agent 'Hupai Jiangnan Xing'. Three agents, 'Yuzhi Heluo', 'Daluo Zhi Yan', and 'Xiangui Xiao Gongdan', were developed and delivered by SenseTime's Large Device team, acting as digital staff for flood-control scheduling, spatial insight, and planning updates. The upgraded model strengthens spatial knowledge and reasoning. The Yunyu Starry Sky Large Model, China's first industry foundation model in planning and natural resources, is built on 13,000 high-quality question-answer pairs and 1.5 billion tokens of corpus. The upgrade improves the model's understanding of urban space and its relationships through enhanced knowledge and reasoning capabilities. To deploy agents, SenseTime built a general platform linking model scheduling, information supply, task execution, and capability accumulation. The flood-control agent 'Yuzhi Heluo' conducts precise zoned and time-specific scheduling for the Suzhou Creek basin by integrating rainfall, water level, and drainage flow data. With forecasts 48 hours or even 2 hours ahead and real-time scheduling simulation, it achieves dynamic water balance across river networks, pipe networks, and neighborhood storage spaces. After deployment, spatial computing, dynamic simulation, and intelligent scheduling efficiency rose by 65%, 70%, and 80%, respectively. The spatial insight agent 'Daluo Zhi Yan' integrates 18 years of high-quality spatiotemporal data, allowing officials to track changes and distribution of urban spatial resources in real time through natural language queries. Data analysis indicators that once took months to develop can now be customized in minutes. The planning management agent 'Xiangui Xiao Gongdan' supports one-click guidance for enterprise requests, crowd-sourced platforms, and rapid review of planning maps to advance urban renewal. Using the FDE co-creation model, business experts and R&D teams work together in a closed loop of problem definition, data and knowledge preparation, agent development, business validation, and iterative feedback. Looking ahead, SenseTime plans to evolve the model into an urban-scale macro world model and expand single agents into coordinated urban agent clusters.