Former top executive of Yitu Technology starts a business, and "Yunjinwei" receives angel round financing led by iFLYTEK | 36Kr exclusive.

Text | Wang Fangyu

Editor | Su Jianxun

36Kr has learned that recently, Ningbo Yunjinwei Intelligent Technology Co., Ltd. (hereinafter referred to as "Yunjinwei") announced the completion of a several million yuan angel round of financing. This round of financing is led by iFLYTEK, and the raised funds will be used for the R & D investment in core technology products and to accelerate the construction of the market ecosystem.

Yunjinwei was established in June 2021 and is a technology company focused on developing an embodied intelligent agent operating system. The company is committed to developing an AI brain for embodied intelligence to help enterprises apply large model technology to edge computing and realize the low-cost intelligent transformation of systems and devices.

In terms of the team, before founding Yunjinwei, the company's founder and CEO, Dr. Wang Wenyi, served as the general manager of the Hardware Division of Yitu Technology, one of the "AI Four Little Dragons". The co-founder, Dr. Zhou Chang, once served as the head of the research and development of the urban brain visual large model of Alibaba DAMO Academy.

Other members of the core team have many years of experience in industries such as AI, chips, and OS. The team's project has won in the Yaojiang Talent Program and the Yongjiang Talent Project in Ningbo City, and has received tens of millions of yuan in government subsidies.

Under the wave of large models, the traditional computer vision (CV) is undergoing a huge transformation. The introduction of multimodal large model technology has greatly improved the recognition efficiency and ability of vision, and is expected to accelerate its penetration process in various industries.

In the view of the founder, Wang Wenyi, in industrial production, the demand for identifying target objects in different industry scenarios is extremely diverse and fragmented. In the era of computer vision, there are many algorithmic needs that cannot be covered and satisfied, such as garbage sorting, quality inspection, operation and maintenance, etc. The emergence of large model technology provides more convenient and lower-cost solutions to these needs.

Aiming at this market gap, Yunjinwei independently developed edge computing devices and pre-trained model tools, and controlled the cost within 10,000 yuan, allowing customers to solve single-point problems in a low-threshold and low-cost way. For example, conducting protective clothing wearing detection at the construction site and dangerous goods transport vehicle detection in the chemical industrial park.

It is reported that up to now, Yunjinwei has enabled enterprise developer customers to achieve commercial landing in multiple scenarios such as energy, water affairs, and transportation, providing products and services to nearly 100 enterprise customers of embodied intelligent agent applications, with a cumulative sales amount of tens of millions of yuan. Among the customers, there are many large-scale technology and industrial enterprises such as China Electronics, Guiyang Rail Transit, SAIC Group, and Shanghai Tunnel.

In terms of products, Yunjinwei has created the embodied intelligent agent operating system Yunjin OS and a series of products equipped with Yunjin OS, such as the software and hardware integrated ZN series embodied edge brain, the VTX series embodied edge all-in-one machine, and the no-development model synthesis and development tool VT-Station, etc.

"We have developed a complete one-stop embodied AI edge computing solution to meet the customers' needs for edge computing terminals, embodied intelligent operating systems, model service platforms, and algorithm generation tool software at one time." Wang Wenyi told 36Kr.

Through free inference computing, no-development model services and algorithm tools, and low-cost edge terminals, Yunjinwei has built a three-level rocket for marketing, which can provide customers with multi-level step-by-step services from zero-threshold entry to deep vertical incubation.

"Our goal is to let every device be equipped with the wings of a large model, so low cost and high cost performance are a major feature of Yunjinwei, especially in the domestic adaptation of AI chips." Wang Wenyi said.

It is reported that the distributed collaborative computing framework VT-Transformer in Yunjin OS has open-sourced the pure C language AI inclusive version, which can be used as a substitute for NVIDIA CUDA architecture to enable domestic AI chips, and has made special optimizations for low and medium-end computing power, which can significantly reduce the computing power cost.

"The original million-level large model deployment solution for N or H cards now only requires a hundred thousand level with the Yunjin OS solution, and the original hundred thousand level visual computing solution now only requires a thousand yuan level, which significantly reduces the industry threshold and helps promote the generalization of downstream AI application scenarios."

Currently, the data raw materials required for AI large models are becoming depleted on the Internet, and whether the Scaling Law is ineffective has become a focus topic in the global technology community. Wang Wenyi believes that in the future, the data raw materials for AI large model training will shift from the digital world of the Internet to the physical world, and from centralized network computing to distributed edge computing and cloud-edge collaboration.

In this context, Wang Wenyi said that Yunjinwei will adhere to the vision of "making AI affordable for every enterprise", helping enterprises complete the training and inference computing required for intelligent agents at an extremely low cost and promoting the inclusive development of AI technology.