NIE 2025 | Yu Youping, President of Zhongguancun KEGIN: How Tier-3 Large Models Become the Core Engine for Industrial Efficiency Improvement

The following are the key points of Yu Youping's speech:

There are four major barriers to the implementation of large models.

• Knowledge barriers, and a profound lack of industry Know-How and business data

   

• Scenario barriers, complex business flows, and adaptation dilemmas with large model usage

   

• Trust barriers: the dual challenges of security and value

   

• Talent barriers, lack of compound talents who understand both the industry and AI, and a gap in product-technology transformation

What kind of vertical large models and applications are needed by the market? In short, when enterprises adopt new technologies including large models, their goals are nothing more than six words: "increase revenue, reduce costs, and improve efficiency." For enterprises, the application of vertical large models ultimately needs to return to its commercial essence, which is to create real value. This requires us to provide professional and in-depth services to achieve a closed-loop scenario. Secondly, what enterprises need is an end-to-end solution, as many intelligent applications need to be integrated with the enterprise's existing information systems. Zhongguancun KEGIN focuses on connecting the entire process of enterprises and customers with intelligence. Thirdly, enterprises also need full-chain services. Through our self-developed Dehu Large Model Platform, we conduct diversified enterprise application development to help customers meet various scenario needs. More importantly, we also provide companion services, achieving a closed-loop empowerment of full-chain services.

Zhongguancun KEGIN - A Leading Large Model Technology and Application Company

Zhongguancun KEGIN was established in 2014. It is a leading enterprise in large model technology and applications, a leader in vertical large models in the field of artificial intelligence, and the only company to have made it onto the "2025 Fortune China's Top 50 Tech Companies" and the "2024 Hurun China's Top 50 AI Companies" list based on its vertical large model advantages. Zhongguancun KEGIN takes the "platform + application + service" three-level engine strategy as its core, focusing on high-value industries and scenarios such as finance, government affairs, industry, automotive, and enterprise globalization. It has created vertical large model solutions covering multiple industries and has served over 2,000 leading enterprises in various industries, including 50% of China's top 100 banks, well-known central and state-owned enterprises, automakers, manufacturing companies, etc.

Build an industry-leading Dehao Large Model Platform

Based on the 'Platform + Application + Service' three-tier engine strategy, Zhongguancun KEGIN has independently developed the 'Dehuo Large Model Platform', covering all-linkage capabilities for large model development and application, including computing power, data, models, and intelligent agents.

The platform boasts three capabilities: unified computing resource scheduling, one-stop model training and deployment, and rapid application construction. It supports mainstream domestic and international computing resource platforms and freely combines with mainstream basic models. It is one of the first platforms in China to connect to the MCP protocol. Unified computing resource scheduling enables shared scheduling of GPU computing resources and unified management of models, including all cutting-edge large model technologies such as open-source models and open MaaS platforms, while also taking into account domestic innovation and adaptation of information technology. In terms of model training and deployment, it supports integrated training and deployment, quantization compression, and modular deployment.

Based on the Dehao Large Model Platform, we have also developed an enterprise-level intelligent agent development platform, which includes more than 200 AI capability components and over 100 ready-to-use industry-specific intelligent agents. It deeply integrates industry Know-How to help enterprises quickly build their own intelligent agents. The AI Agent product matrix focuses on enterprise-level scenarios, covering all links such as marketing, customer service, operations, and office work, such as intelligent outbound calls, intelligent customer service, intelligent coaching, intelligent quality inspection, and intelligent insights. At the same time, it also focuses on vertical scenarios in key industries such as finance, government affairs, industry, automotive, and retail, and provides a large number of intelligent agents in specialized fields.

Security and compliance are also crucial. In recent months, several major companies have open-sourced intelligent agent platforms and large model development platforms. Some are worried that vertical large model platform companies may lose their market due to this, but I believe this is actually a good thing. After more enterprises start using open-source platforms, they will realize that companies need more professional, secure, and compliant platforms. In fact, in any wave of open-source development, there is still significant room for platforms that can truly meet enterprise-level needs.

Rich practical implementation and industry benchmark cases

Zhongguancun KEGIN has collaborated with Ningxia Jiaojian to create the country's first large-scale transportation infrastructure model, "Lingzhu ZhiGong," and built vertical scenario intelligent agents within the industry, such as knowledge Q&A, writing assistants, intelligent generation of accounting reports, and intelligent bidding. By deeply managing and training tens of thousands of industry standard documents through the Dehuo Large Model Platform, the accuracy rate of its professional knowledge answers has been significantly increased by 40% compared to general models, truly becoming an intelligent advisor trusted by engineers.

Zhongguancun KEGIN, in collaboration with leading securities firms, has created an intelligent investment advisor system that monitors the latest market dynamics and data, matches financial products with investment portfolios, generates professional marketing scripts for advisors, and meets the strict data output format and language accuracy requirements of the securities industry, accelerating advisor services.

Intelligent coaching in the securities industry is another typical scenario. Although machines cannot simply replace humans, they can use large models to simulate high-difficulty customer conversations, enabling sales managers and customer service personnel to quickly improve their business capabilities during practical drills. The overall training efficiency can be increased by up to 60%.

For the automotive industry, Zhongguancun KEGIN provides a full-process large model solution that spans pre-sales marketing, in-sales reception, and after-sales service, redefining the automotive marketing experience. The large model insight quality inspection platform we have built can comprehensively monitor and analyze sales behaviors in scenarios such as outbound marketing, in-store reception, and car test drives, accurately detecting the standardization, completeness, and effectiveness of services. Through in-depth customer insight, we can accurately judge the customer's intention to purchase a car, significantly improving the core conversion rate from test drives to signing.

Currently, Zhongguancun KEGIN is also vigorously expanding overseas markets to help Chinese enterprises build global service capabilities when going global. LeOrange IMOU is a smart home brand targeting the consumer end of the global civilian smart IoT market. Its products are available in over 100 countries around the world, with more than 45 million users. Based on our self-developed Instadesk multilingual intelligent customer service system, we have built an intelligent global customer contact center for LeOrange, meeting privacy protection regulatory requirements worldwide, and achieving unified access to multiple channels such as global hotlines, online customer service, social media, and e-commerce internal messages. Relying on AI real-time translation, it can help the customer service team break down communication barriers and assist Chinese enterprises in achieving a dual improvement in global service efficiency and experience when going global.