A：Since 2013, Haizhi Technology has focused on the research and development of knowledge map technology. Later, it was found that the map database technology at that time could not give full play to the value of the map. Therefore, it began to independently develop the map database capability in 2019 and made a breakthrough in 2023, beating the world record at that time. With the rise of large language models, Haizhi combines mapping technology with large language models to achieve the feasibility of industrial-level AI landing applications.

A：Haizhi Technology is the first company in China to apply graph fusion technology to industry-level AI solutions, and ranks first among AI intelligence providers with graph as the core. Our self-developed distributed cloud-native graph database system broke the world record at that time by 45% in 2023.

A：Since 2023, the smart body business of Haizhi Technology has achieved rapid growth, with a growth rate of 1567 percent. Although the growth rate has slowed down, the annual growth rate of more than 50% is still expected. At present, Haizhi Technology has more than 400 customers and will be profitable at the operational level in 2024.

A：Haizhi Technology has the country's top expert Academician Zheng Weimin as the chief scientist, and he jointly established a research institution in 2021, so that the company has more high-level talent cooperation opportunities in the field of graph computing. In addition, the company can adapt to the mainstream model system at home and abroad, and is the national key research and development task of the undertaker, participated in a number of important projects.

A：Haizhi Technology's customer base covers a wide range, including but not limited to land, non-land, finance, energy, telecommunications, transportation and other fields, especially in large enterprises or service targets with complex data and deep accumulation of industrial energy consumption. 70% of state-owned and joint-stock banks have become customers, and the coverage rates of provincial public security, power grid and other fields have also exceeded 60% and 30% respectively.

A：The founders of Haizhi Technology include Ren Zong (one of Baidu's early founding veterans, who hatched many companies after financial freedom, including Haizhi), Yang Na (former host of CCTV and founder of public relations company) and Academician Zheng Weimin (academician of Chinese Academy of Engineering, leader of computer system structure). Academician Zheng joined Haizhi in 2019 as the company's chief scientist.

A：The security problems encountered by OpenAI after the fire explosion include permission problems, rule problems, and application problems in specific scenarios such as the venue in the morning and afternoon. These issues pose a huge challenge for C- end users. In the more complex industrial end landing, these problems are more prominent, leading users to seek to uninstall.

A：SO pic and OpenAI recruit FD engineers in order to really deploy intelligent bodies in the front line of enterprises. These engineers play the roles of front-end engineers, product managers and project managers while deploying intelligent bodies. They deeply integrate industrial energy consumption with AI systems, proving that the application of large language models on the B side cannot be separated from the combination with industrial energy consumption.

A：The first change was the prop project, which improved the model effect by improving the cue words; however, it was found that the structural ability and memory limitations of the model itself led to limited effect, and then developed to the second round of context project, which enhanced the model's memory and background knowledge by establishing a working document database. But in the end, it was found that only providing static documents and background knowledge still could not make the model have the ability to work, so the third generation of harness engineering emerged, by constructing the guidance system of the intelligent body, so that it can understand and follow the knowledge, rules and tacit rules of the specific field, and play the correct role.

A：The importance of harnass engineering is that it provides a set of constraints for the operation of the intelligent body, helps the intelligent body understand the semantic knowledge map, forms the understanding of the professional field, and arranges the tool call and task arrangement reasonably, and solves the long-term logical reasoning problem. At the same time, it emphasizes the deep integration with industrial energy consumption, so that smart energy can be effectively applied and made decisions based on the know-how accumulated by practical experience, not just based on the output of theoretical knowledge.

A：In the grid company scenario, when there is no ontology-based harnass system, the AI can only provide basic data analysis information, such as increased failure rate and increased repair time. Under the perfect harnass system, the AI can not only identify the risk of family defect diffusion, but also predict the future repair needs, evaluate the supplier delivery cycle, and recommend timely maintenance to avoid potential failure, so as to play the value of the intelligent body more fully.

A：If AI want to become a real digital employee, they need to have the corresponding knowledge level and reserve ability, not just provide information query function. Through the ontology system or hard system, AI can understand and correlate the causal relationship between different phenomena, such as equipment failure caused by procurement, maintenance requirements changes, etc., so as to achieve more complex logical judgment and action deployment.

A：AI that do not have an ontology or hard system may give some basic and uninstructive responses, such as changes in the landscape rate due to the increase in new energy, an increase in the frequency of energy storage calls, and an increase in grid regulation. However, this answer cannot directly guide specific actions and still depends on expert advice.

A：The ontology with industry knowledge can reveal that after the increase of new energy output exceeds a certain threshold (such as 15%), a series of chain reactions may be triggered, such as local reverse power flow, mismatch of energy storage strategy leading to loss of regulation ability of power grid, insufficient response of industrial load, etc., and point out the possible follow-up action points brought by these problems.

A：In the face of large-scale complex correlation analysis at the level of tens of billions or even hundreds of billions, the advantage of graph computing is that its computing performance is 1000 times faster than that of traditional relational databases. The graph calculation can be applied to various fields such as power grid, rail transit, social network and so on. It provides a new way to understand the world, and expresses things and their related relationships through the clues of the point-to-point relationship.

A：Relational databases are suitable for expressing simple one-to-many or many-to-many relationships, but when extremely complex multi-layered nested relationships are involved, such as when the company's equity structure penetrates to the underlying entity enterprise, relational databases are no longer effective. At this point, the knowledge graph can fully and dynamically describe the complex relationship network with a three-dimensional data structure, drive tool calls and predict potential chain reactions.

A：Haizhi is not only good at static knowledge organization and mapping, but also pays more attention to the impact of dynamic changes on the whole knowledge system. It has automatic composition technology and can quickly complete the association construction between structured data and the integration and association of structural data and unstructured data.

A：Before large language model technology, building a very large-scale knowledge graph is expensive and requires a large number of industry experts. Now, with the help of large language models and industry understanding, it is possible to achieve automated composition, so that the knowledge map plays a greater role in the intelligent body system, such as task planner, context assembler, executive coder and other components, to provide the intelligent body with the decision-making ability of automatic driving.

A：Intelligent body construction integrates three underlying capabilities: first, a data knowledge platform called DMC, which constructs complex data of enterprises to form AI ready data; The second is the industry ontology of an industry, which integrates and understands the semantic definitions of different departments in the same company based on the same data. Finally, graph-mode fusion technology adapts these capabilities to various size models. There are two reasons why Haiintelligent has done a good job: first, we have accumulated more than ten years of experience and a large number of cases in related industries, and have a deep understanding of the operation mechanism of different industries; second, in the field of financial supervision and public safety that we first entered, the data processed is mainly exogenous data, which gives us advantages in the understanding and application of cross-industry data, and has a leading graph database technology and a number of national standard-setting experience.

A：Haizhi is based on the services of the public security industry. Since public security data mainly comes from exogenous data, and these data are often one of the required data sources in the subsequent manufacturing industries such as finance and power, we can quickly switch and apply it to other fields. At the same time, through in-depth service to various industries, we have cultivated the basic understanding of cross-industry data, so as to realize the construction of knowledge base to knowledge ontology, so that the big meta-model plays a role in the action AI, and connects the optimal legal mechanism for the execution of the whole data business.

A：Qianchuan Xiaozhi is a professional vertical intelligent body that combines large model technology and graph computing technology, and can independently complete task planning, tool invocation and value achievement delivery according to user intention. For example, in the case investigation scene, the investigation and judgment ideas are generated according to the alarm information and map research and judgment knowledge, and through the execution of a series of instructions, the relevant information of the suspect is automatically analyzed to form a case investigation report, which greatly improves the efficiency and accuracy of case detection.

A：The AI agent precipitates the impact of each action through interaction with different departments and functions, and its wisdom increases as the application of the model increases. Just as people learn from experience, AI can adjust their strategies based on historical data and real-time conditions to improve the accuracy and efficiency of their decisions.

A：Without such an intelligent system, companies will make chaotic decisions when faced with multi-objective trade-offs and multi-constraint collaboration. For example, when the tax task is not completed, it may be caused by the poor performance of the regulated enterprises or the insufficient number of new enterprises, and these reasons may be distributed in the data system of the tax bureau, the investment promotion department and the industrial and commercial cooperation. Only through a perfect intelligent system can we penetrate and integrate the tasks and data of various departments, form a global view, and make effective decisions based on this.

A：The intelligent system is composed of multiple sub-agents, which can handle complex multi-objective and multi-constraint environments, and has the ability of comprehensive analysis and multi-sector coordination. Take the AI commander as an example, it can monitor the passenger flow of the city's transportation hub in real time, and make predictions and decisions based on various real-time information (such as people flow, traffic flow, traffic light status, train frequency, etc.), so as to effectively dispatch resources, such as adjusting traffic light duration, directing traffic or adjusting train incoming frequency, etc.

A：First of all, super-large enterprises in the policy-driven new industry landing has a huge ability to pay, and the established industry ontology capabilities have the possibility of extrapolation and generalization. Second, cutting in with an industry leader ensures that customers with adequate budgets and strong needs are available, thus building the potential to serve all customers in the industry. Although such customers tend to privatize deployment and project cooperation at present, in the long run, with the development of business model and the growth of market demand, it is expected to realize the business model of pay according to usage.

A：The company has already achieved break-even in 2024, while the 2025 target is affected by a number of financial treatment factors. Although the current investment in the field of large profits is not the primary goal, but the company's high gross margin and research and development investment is underestimated, with AI demand spurt growth, real revenue growth is expected to be much higher than expected. With the rapid iteration of large model capabilities and the growth of the application side, it is expected that the growth of the application side will accelerate after 2028 to 2029, further moving away from the existing situation.