Recommendation
Huawei AI Native Application Engine Revealed: A one-stop platform helps enterprises achieve intelligent application transformation, covering everything from model management to knowledge engineering.
Core content:
1. Analysis of the five core component architectures of Huawei AI Native Engine
2. Innovative evaluation and governance mechanism of the model center
3. Key technical breakthroughs in knowledge engineering and agent orchestration
Yang Fangxian
Founder of 53A/Most Valuable Expert of Tencent Cloud (TVP)
In recent years, AI big model technology has made breakthrough progress. Its powerful language understanding, generation capabilities, and wide adaptability to application scenarios have completely changed people's perception of AI. Big models can process massive amounts of data, learn complex patterns and knowledge, and provide powerful intelligent support for various applications, ushering in a new era from traditional AI to generative AI.
As big model technology matures, AI native applications emerge. Unlike traditional applications, AI native applications use AI technology as the core driving force and deeply integrate AI capabilities from the beginning of design to achieve more efficient and intelligent user experience and business value. AI native applications have shown great potential in content creation, intelligent customer service, data analysis, decision support and other fields, becoming the key to digital transformation and innovative development of enterprises.
Huawei AI Native Application Engine is a one-stop enterprise-specific native intelligent application development platform, designed to help enterprises select, use, and manage big models from an application perspective, and support AI application innovation. It provides a complete tool chain for enterprise-specific big model development and application development, covering data preparation, model selection/tuning, knowledge engineering, model orchestration, application deployment, application integration and other capabilities , helping enterprise customers integrate their own big model capabilities into their own business application links or external application services, and achieve competitive transformation from traditional applications to intelligent applications.
01 Analysis of AI Native Architecture
1.Overview of the overall architecture
The architecture of Huawei's AI native application engine is exquisitely designed, with multiple key components working together to form an organic whole. Its core components include the model center, knowledge center, agent orchestration center, AI trusted governance, and AI asset library . These components work together to provide all-round support for the development, operation, and management of AI native applications.
2. Analysis of key components
The model center is like the "brain center" of the entire engine, responsible for key responsibilities such as model evaluation, scheduling, observability statistics, and governance.Its core component, the model gateway , plays a vital role. It can centrally host the industry's mainstream large models, commercial large models, and self-built models, and realize the centralized management and scheduling of models from various sources and for different purposes. Through the model gateway, developers can quickly and easily inject existing or third-party model assets for pre-integration verification so that upper-level applications can call them efficiently.In addition, the Model Center has also introduced an ABM-based metadata management mechanism , which builds a clear and specific management list based on key information such as model type, specification, source and purpose, effectively ensuring the security and compliance of model assets in business activities.In terms of model evaluation , the Model Center brings together Huawei's rich experience in scenario practice and multi-dimensional data from internal and external ecosystems to build an automated evaluation framework for model scenarios. In view of the particularity and scarcity of B-side enterprise production scenario data, the framework adopts an innovative evaluation method. On the one hand, with the help of automated means based on AI to judge AI, the evaluation model is used to conduct a preliminary evaluation of the generated results; on the other hand, the professional experience and knowledge of experts in related fields are fully absorbed, and the experts manually annotate or judge the results. Through the organic combination of the two, a more accurate and reliable final evaluation result is obtained.At the same time, on the model center gateway node channel, strict compliance filtering and monitoring are performed on the input and output content, and the performance of model reasoning is monitored in real time, providing a comprehensive and objective measurement and tracking mechanism for the back-end model to ensure the stability and reliability of model operation.The Knowledge Center is the "wisdom treasure house" of Huawei's AI native application engine. Through a series of advanced technical means such as prompt word engineering, model retrieval enhancement generation (RAG) engineering, and agent engineering, it is committed to improving the application effect of the model in various scenarios and achieving continuous iterative optimization. Its workflow begins with importing the company's business experience and knowledge into the platform in the form of document assets, and then using advanced data processing technology to convert these original document data into knowledge forms that the model can understand and effectively use.On this basis, the knowledge center provides comprehensive data processing, annotation processing and generation functions , which can perform fine-grained processing of knowledge data according to different business needs and model training requirements, providing solid data support for the efficient operation of the model and intelligent decision-making.For example, in terms of prompt word engineering, the knowledge center can deeply precipitate and intelligently optimize prompt words. By continuously learning and analyzing a large number of actual application cases, it can automatically generate more accurate and efficient prompt words, and guide the model to generate high-quality results that better meet business needs.In the model retrieval enhanced generation (RAG) project, the knowledge center uses multi-channel (multiple knowledge bases + search engines) generation sampling technology, combined with the detection feedback mechanism, to filter out the most valuable information from massive data, providing strong support for the model to generate the best results, and effectively improving the accuracy and reliability of the model's answers in complex scenarios.(3) Agent Orchestration Center
The Agent Orchestration Center is the "command center" for building complex business logic and intelligent applications . Based on the model, it provides powerful capabilities that can effectively connect the company's existing business capabilities, data resources, and other different third-party models, thereby realizing the automated process orchestration of complex business activities.Huawei's AI native application engine has significant advantages in agent orchestration and supports zero-code agent orchestration . Through a simple and intuitive visual operation interface, even non-professional technicians can easily get started and quickly build intelligent application processes that meet business needs.At the same time, for some complex scenarios that require high flexibility and customization, the engine also provides an SDK method, allowing developers to connect different models or AI agents by writing code to complete high-code orchestration. This flexible and diverse orchestration method provides developers with a wealth of choices and can meet the diverse needs of different companies and different business scenarios.In addition, the Agent Orchestration Center is equipped with a rich set of scenario templates, a powerful stream processing engine, and a complete processing mechanism , and has the ability to efficiently schedule and coordinate tens of millions of Agents. It also provides comprehensive support for function programming and code execution, ensuring that when processing complex business logic, system resources can be efficiently utilized to achieve flexible, convenient, and efficient business process automation, providing a solid guarantee for enterprises to create intelligent and personalized application solutions.(4) Trustworthy AI GovernanceIn today's digital age, data security and compliance are the lifeline of enterprise development. As the "security guard" of Huawei's AI native application engine, the AI Trusted Governance Module protects the reliability of model-generated content results and the security of data assets and model assets through built-in platform security mechanisms and capabilities. Based on Huawei's rich practical experience accumulated over a long period of time, the AI Trusted Governance Module integrates trusted tools and methods for the entire process, and can perform refined access control and security protection for data, knowledge bases, Agent applications, and related business processes based on different roles and permission settings within the enterprise.For example, in terms of data security, strict encryption measures are taken throughout the entire life cycle of data collection, storage, transmission and use to ensure the confidentiality and integrity of the data. In terms of model interaction security, by building a model gateway within the enterprise, internal and external models are uniformly accessed and managed, and on this basis, three layers of security isolation belts are carefully built. The first layer establishes a security scoring filter for the content returned by the model, and through general inspection or random inspection, it promptly discovers the potential security risks and risks of the model itself; the second layer establishes an enterprise information security filter to conduct a comprehensive inspection of all model call requests to ensure that no sensitive information is included in the call process; the third layer establishes an enterprise field filter. Taking into account the differences in sensitive words and security control strength in different fields, each field can set and adjust security policies based on this mechanism to better isolate the company's key information in the internal security environment, effectively prevent data leakage and abuse risks, and ensure the stable operation of AI native applications in a safe and compliant environment.The AI asset library is a valuable "wealth warehouse" accumulated by Huawei's AI native application engine in long-term development activities. It brings together various scene models, data resources, knowledge and experience, prompt word templates, agent applications, and business activity assets that have been accumulated during the development process, aiming to achieve the maximum reuse of these assets, thereby significantly improving the efficiency and quality of subsequent development work. The asset center has built-in a large number of data assets from industry partners and actual industry practices. These data have been carefully sorted and annotated and have extremely high value and practicality. At the same time, in the API Hub, API assets with many industry capabilities are pre-packaged, covering various scene API Kits commonly used in production and life such as truck logistics, weather query, and taxi services . These API assets can be easily called by large models or agents, greatly enriching the functions and service scope of the application. In addition, AI native assets based on large models, such as agent development templates, scene models, knowledge assets, etc., provide developers with ready-made solutions and reference frameworks, effectively reducing the difficulty and complexity of development, helping developers to quickly complete application innovation and accelerate the process of enterprise intelligent transformation.02 Detailed explanation of functions and features
1. Model access and management
Huawei's AI native application engine has strong compatibility. At this stage, it has pre-installed 30+ mainstream large models in the industry and continues to actively expand, with the goal of achieving access capabilities for hundreds of models and thousands of forms. Whether it is an open source large model, a leading model in the commercial field, or a self-built model customized and developed by an enterprise according to its own business needs, it can be seamlessly accessed on the engine platform. Through the integration of multiple models, enterprises can flexibly select the most suitable model according to different business scenarios and needs, give full play to the advantages of each model, and provide more choices and stronger technical support for business innovation.(2) Unified model governanceIn order to effectively manage the numerous models connected, Huawei AI native application engine provides comprehensive unified model governance functions. This includes unified aggregation of models, centralized management of scattered model resources, and facilitating overall control and scheduling by enterprises; unified scheduling, intelligent allocation of model resources based on business load, model performance, user needs and other factors, to ensure that the model can provide services efficiently and stably during operation; unified evaluation, using scientific and reasonable evaluation indicators and methods to objectively evaluate the performance of different models in various business scenarios, and provide data basis for enterprises to select the best model; and unified distribution, accurately and timely distributing the model results after evaluation and screening to the corresponding applications or users to ensure the smooth progress of business processes. Through these unified governance measures, enterprises can better manage the complex model ecosystem, improve model usage efficiency, and reduce management costs.(3) Model routing and optimization
Model routing is a key intelligent function of Huawei's AI native application engine. It can automatically analyze and determine the most suitable model for different business requests based on the evaluation results of the model, and accurately route the request to the corresponding model for processing. This intelligent model selection mechanism can give full play to the strengths of each model and provide higher-quality and more efficient services for upper-layer applications. For example, for some text creation scenarios that require high language generation quality, model routing will give priority to models that perform well in language generation; while for business scenarios that require complex data analysis and reasoning, requests will be directed to models that are good at such tasks. At the same time, Huawei's AI native application engine also has model optimization capabilities, which can continuously adjust and optimize the model according to the feedback data of the model during actual operation, improve the performance and accuracy of the model, and make it better adapt to changing business needs. Through continuous model routing and optimization, enterprises can significantly improve the overall effect and user experience of AI applications without increasing too much resource investment.2. Knowledge Engineering Support
(1) Prompt word management
Prompt words play a key role in guiding models to generate accurate and relevant results in AI applications. The prompt word management function of Huawei AI native application engine provides enterprises with powerful prompt word precipitation and optimization tools. In actual business applications, enterprises can collect and organize effective prompt words used in various scenarios and store them in the prompt word library of the engine. With the continuous development of business and the continuous accumulation of data, the prompt word library will continue to be enriched and improved. At the same time, the engine uses advanced machine learning algorithms to conduct in-depth analysis and mining of the data in the prompt word library, and automatically identify the prompt word patterns and rules that can guide the model to generate high-quality results. On this basis, the prompt words are intelligently optimized to generate more accurate and efficient prompt words to improve the quality and accuracy of the model's answers in various business scenarios. For example, in the intelligent customer service scenario, by optimizing the prompt words, the model can understand the user's questions more accurately and provide solutions that better meet user needs, greatly improving customer service efficiency and user satisfaction.(2) High-quality data preparation and management
Data is the foundation of AI applications, and high-quality data is crucial to improving model performance. Huawei's AI native application engine provides comprehensive high-quality data preparation and management functions, covering 6 types of rich data sets, including text, images, audio, video, structured data, and semi-structured data, which can meet the diverse data needs of different industries and business scenarios. In the data preparation stage, the engine provides a series of professional data processing tools, such as data cleaning, denoising, labeling, conversion, etc., to help enterprises process raw data into high-quality data formats suitable for model training and application. At the same time, in order to further improve the efficiency and accuracy of data labeling, the engine introduces advanced data labeling enhancement technology. Using AI-assisted labeling tools, it can automatically identify and label key information in the data, reduce the workload and errors of manual labeling, and improve the quality and efficiency of labeling. In terms of data management, the engine provides complete data storage, retrieval, version control and other functions to ensure that the data assets of enterprises can be managed and utilized safely and effectively, and provide solid data support for the continuous optimization and innovation of AI applications.(3) Model-enhanced search (RAG)
Model-augmented search (RAG) is a core technology of Huawei's AI native application engine to improve model performance and answer accuracy. When faced with complex business problems and diverse user needs, a single knowledge base or model often cannot provide comprehensive and accurate answers. RAG technology fully utilizes the information advantages of multiple data sources by adopting a multi-channel (multiple knowledge bases + search engine) sampling method. When a user asks a question, the engine first retrieves relevant information from multiple knowledge bases, combines search engines to obtain the latest external knowledge, and then inputs this multivariate information into the model for comprehensive analysis and processing. In the process of generating answer results, RAG technology will also evaluate and adjust the generated results in real time based on the detection feedback mechanism. If the results are found to be inaccurate or incomplete, the relevant information will be automatically retrieved and analyzed again, and the results will be generated again until the best answer result is obtained. In this way, RAG technology can significantly improve the accuracy and reliability of the model's answers in complex scenarios, providing users with more comprehensive and high-quality services. For example, in an enterprise's intelligent question-and-answer system, RAG technology can integrate multiple sources of information, such as the enterprise's internal product knowledge base, technical document library, and external industry information, to provide users with richer and more accurate answers, help users quickly solve problems, and improve the enterprise's service quality and competitiveness.The knowledge flywheel is an innovative mechanism for Huawei's AI native application engine to achieve continuous learning and self-optimization. In the daily business operations of enterprises, business personnel will continuously accumulate a large amount of incremental data, such as user feedback information, business operation records, new market data, etc. At the same time, users will also generate various feedback in the process of using AI applications, such as satisfaction evaluation of the answer results, new types of questions raised, etc. Huawei's AI native application engine has built a knowledge flywheel system through a series of technical means such as multi-tool pre-integration, mainstream model integration, business application information extraction plug-in, intelligent application feedback plug-in, and automatic scheduling. The system can collect, organize and analyze the incremental data and user feedback information accumulated by these business personnel in real time, and refine them into enterprise knowledge. Then, the newly generated enterprise knowledge is continuously supplied to the large model for multiple rounds of cyclic learning, so that the model can continuously update and optimize its own knowledge system and achieve daily iteration. As the knowledge flywheel continues to operate, the intelligence level and application effect of the model will continue to improve, providing enterprises with more intelligent and personalized services, forming a virtuous cycle development model, and helping enterprises to always maintain their competitive advantage in a rapidly changing market environment.3. Intelligent architecture support
(1) Generate Agent applications in minutesHuawei's AI native application engine innovatively implements the function of generating agent applications in minutes, greatly reducing the threshold for developing AI native applications and making agent application development more accessible to all. In the traditional application development model, developing a fully functional intelligent application often requires a lot of time, manpower and technical resources, with a long development cycle and high costs. Huawei's AI native application engine breaks this traditional bottleneck by introducing advanced intelligent architecture and automated development tools. When enterprise users describe their business demands and business scenarios on the platform, the engine can quickly start the intelligent generation process. It will automatically filter appropriate data from the data resource library based on the information provided by the user, call the corresponding tools and algorithms, and mount the matching knowledge base. At the same time, it selects the best prompt word template and generates a complete agent application in a short time in an intelligent and automated way. This "Agent for Agent" development model not only greatly shortens the application development cycle and improves development efficiency, but also enables non-professional developers to easily participate in the development of AI native applications, providing a wider range of technical support and talent base for the innovative development of enterprises.(2) One-stop deployment of cloud, edge and terminal
In order to meet the diverse needs of enterprises for AI application deployment in different scenarios, Huawei AI native application engine provides a one-stop deployment solution for cloud, edge and terminal. AI applications developed by enterprises only need to be built once, and can be deployed, monitored and maintained in a one-click manner on cloud, edge and multiple terminals through the engine platform. In terms of cloud deployment, the engine makes full use of the powerful computing resources and storage capabilities of cloud computing to provide a stable and efficient operating environment for large-scale, high-concurrency application scenarios. Cloud deployment is suitable for applications that require high data processing capabilities and need to interact with a large number of users in real time, such as the company's online customer service system and intelligent marketing platform. In terms of edge deployment, considering that some scenarios have strict requirements on real-time and low latency, such as equipment monitoring and control in industrial manufacturing and real-time video analysis in intelligent security, the engine can deploy AI applications on edge devices close to the data source to reduce data transmission delays and improve system response speed. At the same time, through cloud-edge collaboration technology, the cloud and edge can achieve real-time synchronization and collaborative work of data, ensuring that applications can run stably in different environments. In addition, the engine also provides comprehensive support for some AI applications that need to be used on mobile devices, enabling rapid deployment of applications to various mobile devices, providing users with convenient mobile intelligent services. Through one-stop deployment of cloud, edge, and end, Huawei's AI native application engine provides enterprises with a more flexible and efficient application deployment method, improving their digital operation capabilities and user experience.(3) Hyper-distributed architecture
Huawei's AI native application engine uses a hyper-distributed architecture to achieve high-performance communication and unified management of tens of millions of agents. In large-scale enterprise-level application scenarios, a large number of agents are often required and coordinated to complete complex business tasks. The hyper-distributed architecture breaks the performance bottleneck and resource limitations of the traditional centralized architecture by distributing and managing the system's computing, storage, and communication resources. Under this architecture, each agent can run independently and process some tasks, while being able to interact and share data with other agents in real time through efficient communication mechanisms to achieve collaborative processing of tasks and dynamic allocation of resources. For example, in the intelligent recommendation system of a large e-commerce platform, it is necessary to process a large amount of user browsing behavior data, product information data, and various marketing activity data at the same time. Through the hyper-distributed architecture, tens of millions of agents can analyze and process this data in parallel, quickly generate accurate product recommendation lists based on the personalized needs of users, and greatly improve the response speed and accuracy of the recommendation system. At the same time, the hyper-distributed architecture also has strong scalability and fault tolerance, and can easily cope with the continuous expansion of business scale and dynamic changes in system load. Even if some agents fail, the system can ensure the normal operation of the entire application through intelligent failover and task reallocation mechanisms, providing solid and reliable technical support for the company's key businesses.03 Application Case Display1. Financial field: intelligent risk control and customer service
In the financial industry, risk control and customer service are core business links. A large bank introduced Huawei's AI native application engine to build an intelligent risk control and customer service system.
In terms of intelligent risk control, the engine integrates a variety of financial data, including structured data such as customer transaction records, credit history, asset status, and unstructured data such as social media information and news and public opinion. Through the model center, it accesses advanced risk assessment models, uses the knowledge center to conduct in-depth analysis and processing of data, and extracts key risk features. When there is a new loan application, the system can automatically assess the customer's risk level, and combined with model-enhanced search (RAG) technology, retrieve relevant cases and risk indicators from multiple knowledge bases, providing a comprehensive and accurate basis for risk control decisions, greatly reducing credit risks.
In terms of customer service, the intelligent customer service system developed based on Huawei's AI native application engine can quickly and accurately understand customer questions through the optimization of prompt words and continuous learning of the knowledge flywheel. Whether it is account inquiries, business consultations or complaints and suggestions, professional answers and solutions can be given in a timely manner. At the same time, the system can also proactively provide personalized financial product recommendations based on the customer's historical behavior and preferences, improving customer satisfaction and business conversion rate.
2. Manufacturing: Intelligent production and equipment maintenance
In the manufacturing industry, Huawei's AI native application engine helps companies achieve intelligent upgrades.
An automobile manufacturer has used the engine to build an intelligent production management system. During the production process, the system collects real-time operating data of production equipment, such as temperature, pressure, and speed, through sensors deployed on the edge. After the data is transmitted to the cloud, it is analyzed by the engine's model center, which combines the preset production model and historical data to monitor abnormal conditions in the production process in real time. Once it is found that the equipment operating parameters deviate from the normal range, the Agent orchestration center will automatically trigger the early warning mechanism and coordinate with relevant departments to handle it according to the preset rules and processes, such as arranging maintenance personnel and adjusting production plans, thus realizing intelligent monitoring and management of the production process.
In terms of equipment maintenance, the knowledge center uses historical equipment maintenance records, failure cases and other data to build an equipment failure prediction model. Through continuous monitoring and analysis of equipment operation data, it is possible to predict possible equipment failures in advance and formulate corresponding maintenance plans, realizing the transformation from traditional post-maintenance to preventive maintenance, effectively reducing equipment failure rates and downtime, and improving production efficiency and product quality.
3. Medical field: auxiliary diagnosis and health management
In the medical industry, Huawei's AI native application engine also shows great application potential.
A medical institution has developed an intelligent auxiliary diagnosis system based on this engine. The system uses a large amount of medical imaging data, medical records, etc., and uses the medical imaging analysis model and disease diagnosis model of the model center to assist in the diagnosis of the patient's condition. When analyzing medical images, the model can automatically identify the characteristics of lesions in the image and compare them with standard cases in the knowledge base to provide doctors with diagnostic suggestions. The knowledge center continuously optimizes the diagnostic accuracy of the model by constantly learning new medical research results and clinical experience.
At the same time, the system can also provide patients with personalized health management services. According to the patient's health data and disease risk assessment results, it can formulate recommendations on diet, exercise, rehabilitation, etc. to help patients prevent diseases and promote recovery. Through Huawei's AI native application engine, medical resources are used more efficiently, diagnostic efficiency and accuracy are improved, and patients are provided with better medical services.
04 Advantages of Huawei's AI Native Application Engine
1. Reduce development threshold and cost
Huawei AI native application engine provides a one-stop development platform and a rich tool chain, which greatly reduces the development threshold of AI native applications. For enterprise developers, there is no need to have a deep AI technical background and complex development experience. Through a simple and intuitive operation interface and templated development method, intelligent applications can be quickly built. For example, the function of generating Agent applications in minutes enables enterprises to convert business needs into practical applications at extremely low cost and in a very short time. At the same time, the rich resources in the AI asset library realize the maximum range of asset reuse, avoid repeated development, and further reduce development costs. For small and medium-sized enterprises, this undoubtedly provides them with a convenient channel to quickly enter the AI field, realize business innovation and digital transformation.
2. Improve application intelligence and efficiency
The engine's powerful model access and management capabilities can select the optimal model according to different business scenarios, and through model routing and optimization mechanisms, give full play to the advantages of each model to provide applications with more intelligent and efficient services. A series of technologies in knowledge engineering, such as prompt word management, model-enhanced search (RAG), knowledge flywheel, etc., continuously improve the model's understanding and response capabilities to business needs, enabling applications to generate more accurate and valuable results. With the support of intelligent architecture, one-stop cloud-edge deployment and ultra-distributed architecture ensure the efficient operation of applications in different environments, meet the company's diverse business needs for real-time, high concurrency, etc., and significantly improve the company's operational efficiency and competitiveness.
3. Ensure data security and compliance
As data security and compliance become increasingly important, the AI Trusted Governance module of Huawei's AI Native Application Engine plays a key role. The trusted tools and methods throughout the entire process, as well as the design of three-layer security isolation zones, strictly protect data assets and model assets from data collection, storage, transmission to use. Enterprises can set different security policies according to their own needs to ensure the security of sensitive information and effectively prevent data leakage and abuse risks, meeting the strict requirements of enterprises in terms of data security and compliance, and providing reliable guarantees for the stable development of enterprises.
