According to an exclusive report by Sina Tech on June 2, Wang Yunhe, former director of Huawei’s Noah’s Ark Lab and core leader of the PanGu large model, has officially launched a startup in the AI agent space after leaving Huawei on March 28 this year. His new company, JiYuan LvDong (registered in Shanghai on April 8), has completed its initial funding round, reaching a post-investment valuation of $100 million. The investors include top-tier domestic venture capital firms and several leading internet companies. The company has already secured stable cooperation orders from several state-owned tech giants, and its first commercial agent product is expected to be released within months. Han Kai, former chief researcher at Huawei’s Noah’s Ark Lab, has joined as CTO. The two have long collaborated on developing algorithms such as GhostNet and AdderNet. The company is currently posting job openings on recruitment platforms for roles including AI algorithms, agent development, and security compliance, with annual salaries ranging from 600,000 to 1,000,000 RMB.
Born in 1991, Wang Yunhe joined Huawei in 2017 as its first intern in the Beijing division and rose to lab director over nine years. During his tenure, he led the development of lightweight architectures such as GhostNet (CVPR 2020) and AdderNet (CVPR 2020 Oral), earning Huawei’s Fourth Top Ten Invention Award. The algorithms have also been applied to China’s FAST telescope, helping discover hundreds of new fast radio burst samples. After becoming director of the Noah’s Ark Lab in March 2025, he quickly launched the PanGu 5.5 series within about three months. The flagship model, Pangu Ultra, uses a MoE architecture with 718B parameters (approximately 39B activated), achieving an 8x improvement in inference efficiency over its predecessor. It has been deeply integrated into the full-chain operations of major automakers including FAW, GAC, Changan, Seres, and BYD. JiYuan LvDong is positioning itself in the enterprise-grade AI agent space for industry implementation, leveraging Wang’s expertise in model compression and lightweight solutions to target AI agent deployment needs in traditional industries such as manufacturing, finance, and automotive, deliberately avoiding the competitive red ocean of general-purpose large models dominated by tech giants.