Application construction method and computing device

Abstract

The embodiment of the invention provides an application construction method and computing equipment. The method comprises the following steps: in response to a construction request for a target application, obtaining first demand information input by a user in a first-stage dialogue; the first demand information is used for describing the capability expected to be realized by the target application; generating a candidate workflow based on the first demand information, and displaying the candidate workflow through a display device; the candidate workflow is used for representing a workflow of the target application; obtaining second demand information input by the user in the second-stage dialogue, and determining a target workflow based on the second demand information; the second demand information is used for describing an improvement direction or a confirmation indication for the candidate workflow; the target workflow accords with the second demand information; and generating a target application based on the target workflow. According to the method, the application can be constructed more accurately according to the requirements of the user.

Description

Technical Field

[0001] This application relates to the field of computing technology, and in particular to an application construction method and computing device. Background Technology

[0002] With the development of artificial intelligence technology, large models are being applied more and more widely in various fields. Large models can understand and generate natural language text, thereby assisting users in solving a wide variety of problems. While large models possess powerful general-purpose capabilities, for users in specialized fields, these general-purpose models often cannot efficiently solve business problems specific to their chosen domain. Therefore, large model applications have emerged. Large model applications are based on large models and integrate specialized knowledge bases, thus effectively assisting users in solving business problems within specific professional domains.

[0003] However, the current construction of large-scale model applications generally relies on a fixed framework, which leads to inaccurate applications that often fail to function properly or effectively solve user problems. Summary of the Invention

[0004] This application provides an application building method and computing device that can more accurately build applications according to user needs.

[0005] In a first aspect, the application building method provided in this application includes: responding to a building request for a target application, obtaining first requirement information input by a user in a first-stage dialogue; the first requirement information is used to describe the capabilities expected to be implemented by the target application; generating candidate workflows based on the first requirement information, and displaying the candidate workflows through a display device; the candidate workflows are used to represent the workflow of the target application; obtaining second requirement information input by a user in a second-stage dialogue, and determining a target workflow based on the second requirement information; the second requirement information is used to describe the improvement direction or confirmation indication for the candidate workflows; the target workflow conforms to the second requirement information; and generating a target application based on the target workflow.

[0006] This application provides an application building method. When a user needs to build a target application, the method first obtains the user's first requirement information from a first-stage dialogue, and then generates candidate workflows based on the first requirement information. Next, it obtains the user's second requirement information from a second-stage dialogue, and then improves or confirms the candidate workflows based on the second requirement information to obtain a target workflow that meets the user's requirements. Finally, it generates the target application based on the target workflow. It can be seen that in the application building process, the solution of this application first uses a first-stage dialogue to comprehensively and accurately obtain the capabilities that the user expects the target application to achieve, thus building an initial workflow. Furthermore, through a second-stage dialogue, it comprehensively and accurately obtains the user's suggestions for improving the initial workflow, and then further adjusts the initial workflow based on the user's suggestions to obtain a target application that more accurately reflects the user's requirements.

[0007] In one possible implementation, generating candidate workflows based on first requirement information includes: employing a first intelligent agent to determine workflow templates matching the first requirement information from a workflow template set; the workflow template set includes multiple workflow templates corresponding to different capabilities; and determining candidate workflows from the candidate workflow templates matching the first requirement information. In the above implementation, by pre-setting multiple workflow templates, workflows that meet user needs can be generated quickly, thereby improving application building efficiency.

[0008] In another possible implementation, determining candidate workflows from candidate workflow templates that match the first requirement information includes: if there are multiple candidate workflow templates that match the first requirement information, using a second agent to obtain third requirement information input by the user in the third-stage dialogue, the third requirement information being used to indicate one or more steps included in the desired target application; using the second agent to determine a score for each candidate workflow template based on the steps in each candidate workflow template and the third requirement information; the score being used to characterize the degree of conformity between the candidate workflow template and the third requirement information; and determining the candidate workflow template whose score meets a first preset condition as a candidate workflow.

[0009] In another possible implementation, when no candidate workflow template matching the first requirement information exists in the workflow template set, the method further includes: using a third agent to obtain an initial workflow based on the first requirement information; and using a fourth agent to configure the functional implementation scheme of each node in the initial workflow based on the first requirement information to obtain candidate workflows. This approach provides a solution for situations where no matching workflow template exists, thereby improving the feasibility and versatility of the proposed solution.

[0010] Another possible implementation involves using a fourth agent to configure the functional implementation scheme of each node in the initial workflow. This includes: for each node in the initial workflow, the fourth agent determines a node template from a set of node templates that matches the first requirement information, and then configures the functional implementation scheme of the node using the node template; the set of node templates includes multiple node templates, and each node template includes a functional implementation scheme for one function. Through this method, attributes at both the functional description and functional implementation scheme levels are configured for each node in the workflow, thereby ensuring that the workflow can run automatically and facilitating subsequent application development.

[0011] In another possible implementation, before displaying the candidate workflow on the display device, the method further includes: using a fifth agent to determine target nodes in the candidate workflow that meet a second preset condition, where the second preset condition indicates that the target node does not need to be displayed on the display device; deleting the target node, or integrating the target node with upstream and / or downstream nodes. Through these methods, the workflow can primarily reflect the business process without displaying excessive internal details, thus improving the user experience.

[0012] In another possible implementation, the method further includes: acquiring initial input information of the target workflow; using a sixth agent to input the initial input information into the target workflow, and sequentially outputting the debugging results of each node; wherein the trigger condition for outputting the debugging results of each node is: the sixth agent receives a debugging instruction from the user, the debugging instruction being used to instruct the target workflow to execute from the current node to the next node. Based on the debugging results of each node, it is determined whether the target workflow is running normally; based on the target workflow, the target application is generated, including: if it is determined that the target workflow is running normally, the target application is generated based on the target workflow.

[0013] In another possible implementation, obtaining the first requirement information input by the user in the first stage of the dialogue includes: using a seventh agent to converse with the user and obtain the user's input content; using the seventh agent to determine whether the input content meets a third preset condition; if it does, determining the input content as the first requirement information; if it does not, using the seventh agent to prompt the user to re-enter; wherein, the third preset condition includes: the input content is associated with the application construction, and the input content can clearly and completely describe the capabilities of the target application. The above implementation communicates with the user in the form of dialogue, thereby enabling a more accurate understanding of the user's needs and the construction of an application that better meets the user's requirements.

[0014] Secondly, this application provides an application building apparatus for executing any of the application building methods provided in the first aspect above.

[0015] Thirdly, embodiments of this application provide a computing device including a processor and a memory; the processor is coupled to the memory; the memory is used to store computer instructions, which are loaded and executed by the processor to enable the computing device to implement the method described in the first aspect.

[0016] Fourthly, embodiments of this application provide a computer-readable storage medium comprising: computer software instructions; when the computer software instructions are executed in a computing device, they cause the computing device to implement the method described in the first aspect.

[0017] Fifthly, embodiments of this application provide a computer program product that, when run on a computing device, causes the computing device to execute the steps of the relevant method described in the first aspect above, so as to implement the method of the first aspect above.

[0018] The beneficial effects of the second to fifth aspects mentioned above can be referred to the corresponding description of the first aspect, and will not be repeated here. Attached Figure Description

[0019] Figure 1 This application provides an architecture diagram of an application building system.

[0020] Figure 2 This is a schematic diagram of an application scenario provided by an embodiment of this application;

[0021] Figure 3 This is a schematic diagram of the hardware structure of a computing device provided in an embodiment of this application;

[0022] Figure 4 A flowchart illustrating an application construction method provided in an embodiment of this application;

[0023] Figure 5 This is a schematic diagram illustrating the composition of an information-gathering intelligent agent provided in an embodiment of this application;

[0024] Figure 6 A schematic diagram of the composition of a workflow provided in an embodiment of this application;

[0025] Figure 7 A schematic diagram of an interface provided in an embodiment of this application;

[0026] Figure 8 A flowchart illustrating a process adjustment provided in an embodiment of this application;

[0027] Figure 9 A schematic diagram of a debugging process provided for an embodiment of this application;

[0028] Figure 10A schematic diagram of a template matching process provided for an embodiment of this application;

[0029] Figure 11 A schematic diagram illustrating a node configuration process provided in an embodiment of this application;

[0030] Figure 12 A schematic diagram illustrating another workflow provided in an embodiment of this application;

[0031] Figure 13 This is a schematic diagram illustrating the operation of a business judgment intelligent agent provided in an embodiment of this application;

[0032] Figure 14 A flowchart illustrating a workflow generation method provided in this application embodiment;

[0033] Figure 15a A schematic diagram showing another interface provided in an embodiment of this application;

[0034] Figure 15b A schematic diagram illustrating yet another interface provided in an embodiment of this application;

[0035] Figure 15c A schematic diagram illustrating yet another interface provided in an embodiment of this application;

[0036] Figure 16 This is a schematic diagram illustrating the complete process of application construction provided in an embodiment of this application. Detailed Implementation

[0037] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0038] It should be noted that in the embodiments of this application, the words "exemplarily" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplarily" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplarily" or "for example" is intended to present the relevant concepts in a specific manner.

[0039] To facilitate a clear description of the technical solutions of the embodiments of this application, the terms "first" and "second" are used in the embodiments of this application to distinguish the same or similar items with essentially the same function and effect. Those skilled in the art can understand that the terms "first" and "second" are not intended to limit the quantity or execution order.

[0040] The following is a brief explanation of the technical terms used in the embodiments of this application:

[0041] 1. Large Language Model (LLM): A type of artificial intelligence model that uses deep learning algorithms to learn language patterns and structures through training on a large amount of text data, thereby enabling it to understand and generate natural language text.

[0042] 2. Intelligent Agent: This refers to an intelligent system built upon a large model, capable of autonomously understanding tasks, planning steps, invoking tools, and achieving complex goals. Through interaction with the large model, it combines natural language understanding, reasoning abilities, and external tools to achieve problem-solving capabilities similar to humans.

[0043] 3. Enterprise knowledge base: An internal system or platform used by an enterprise to store, manage and share knowledge resources. It stores various documents of the enterprise, such as policies, processes, manuals, training materials, etc., and records the experience and lessons learned by the enterprise in different projects and tasks, which are used to help employees solve problems encountered in their work.

[0044] 4. Prompt: This refers to the input text used to guide or stimulate the large language model to perform a specific task. Its purpose is to help the large language module model understand the type of task the user wants to perform or the required output format.

[0045] 5. Retrieval-augmented generation (RAG): This technology allows users to retrieve answers relevant to their questions from a pre-configured enterprise knowledge vector database. These answers, after being concatenated with prompts, are then fed into a larger model to help it better answer user questions.

[0046] This application provides an application building method. When a user needs to build a target application, the method first obtains the user's first requirement information from a first-stage dialogue, and then generates candidate workflows based on the first requirement information. Next, it obtains the user's second requirement information from a second-stage dialogue, and then improves or confirms the candidate workflows based on the second requirement information to obtain a target workflow that meets the user's requirements. Finally, it generates the target application based on the target workflow. It can be seen that in the application building process, the solution of this application first uses a first-stage dialogue to comprehensively and accurately obtain the capabilities that the user expects the target application to achieve, thus building an initial workflow. Furthermore, through a second-stage dialogue, it comprehensively and accurately obtains the user's suggestions for improving the initial workflow, and then further adjusts the initial workflow based on the user's suggestions to obtain a target application that more accurately reflects the user's requirements.

[0047] The embodiments provided in this application will now be described in detail with reference to the accompanying drawings.

[0048] Figure 1 This is a schematic diagram of the architecture of an application building system provided in an embodiment of this application. The system achieves a fully automated process from user requirement collection to application generation through the collaborative work of multiple modules, and relies on global variables to achieve data sharing and state synchronization among the modules. Here, global variables refer to variables that can be accessed and used by every module in the application building system.

[0049] 1. System Architecture and Data Flow

[0050] The application build system operates on global variables, which store structured data generated by each module. This structured data is crucial for building the application, including:

[0051] The application name (application name = []), functional requirements (application function = []), sample data (example = []), and operation process description (application operation process = []) obtained by the information collection module are all user inputs.

[0052] The workflow generated by the template matching module or the workflow generation module includes nodes (nodes=[]) and edges (edges=[]), which constitute the topology of the application logic.

[0053] The node configuration information (configs=[]) obtained by the configuration generation module includes the specific implementation scheme of each node in the workflow, which can support the operation of the application.

[0054] 2. Core Module Function Description

[0055] (1) Information gathering module (also known as information gathering agent, referred to as the seventh agent below)

[0056] This module is responsible for interacting with users using natural language, extracting key information (such as application name, functional requirements, workflow, etc.) from the user's input. After each round of interaction, it checks whether the collected key information meets the application building conditions. If the conditions are met, the key information is stored in a global variable, and the template is triggered to begin working. If the conditions are not met, it continues to interact with the user using natural language, guiding the user to supplement the missing information. Here, the application building conditions can be understood as the clarity, richness, and completeness of the key information. When the key information meets the clarity, richness, and completeness requirements for application building, it is more conducive to building an application that meets the user's needs.

[0057] (2) Template matching module (also known as template matching agent, referred to as the first agent below)

[0058] The template matching module is used to match templates that match the key information from the workflow template set. If the match is successful, the matched template will be used as the initial workflow.

[0059] (3) Workflow generation module (also known as workflow generation intelligent agent, referred to as the third intelligent agent in the following text)

[0060] If there is no matching template in the workflow template set, the workflow generation module will dynamically construct the application's initial workflow based on the key information input by the user.

[0061] (3) Node configuration module (also known as node configuration agent, referred to as the fourth agent below)

[0062] When a workflow is generated by a workflow generation module, the specific implementation details (functional implementation schemes) of each node in the workflow are lacking. Therefore, a node configuration module retrieves suitable functional implementation schemes from the node module set and configures each node in the workflow, thus defining the functional implementation scheme for each node. The node template set includes multiple node templates, and each node template contains the functional implementation scheme for one function.

[0063] (4) Application debugging module (also known as application adjustment agent, referred to as the sixth agent below)

[0064] Perform node-by-node testing on the generated workflow to verify that the inputs / outputs meet expectations. During debugging, return the execution results of each node and the final running status of the entire workflow for user or system optimization reference.

[0065] (5) Process Adjustment Module (also known as Process Adjustment Intelligent Agent)

[0066] Users can dynamically modify workflows through natural language interaction. The adjusted information is recorded to global variables, triggering the regeneration of workflow and node configurations.

[0067] (6) Application generation module (also known as application generation agent)

[0068] Add metadata (such as application name, version, etc.) to the debugged workflow and package it into a deployable application.

[0069] It should be noted that, Figure 1 This is just one example; application building systems can also include more or fewer agents.

[0070] Figure 2 A schematic diagram illustrating an application scenario provided by an embodiment of this application is shown. For example... Figure 2 As shown, it includes a terminal device 200 and a server 210. The application building method provided in this embodiment can be executed by the terminal device 200.

[0071] The terminal device 200 and the server 210 can interact via a communication network. This communication network can be Ethernet, wireless access network, wireless local area network (WLAN), etc.

[0072] Among them, terminal device 200 is a device with interface display function.

[0073] Optionally, the terminal device 200 can be a personal computer, smartphone, tablet computer, e-book reader, portable computer, or other similar device.

[0074] Server 210 deploys a large language model, providing an interactive interface for the large language model to terminal device 200, and allowing terminal device 200 to obtain descriptive information about the application the user needs to build. Then, server 210 uses the large language model to understand the descriptive information, constructs the application workflow, and displays it to the user through terminal device 210. The user can adjust the workflow according to their needs.

[0075] For example, terminal device 200 can access server 210 through a browser to display the interactive interface of the large language model.

[0076] Optionally, the aforementioned server 210 can be a cluster of one or more servers. Figure 1 Each intelligent agent in the process can be deployed on the same server or on different servers. This application does not specifically limit this.

[0077] In terms of form, the server mentioned here can be a blade server, a high-density server, a rack server, or a full-rack server; in terms of function, the server can be a general-purpose server, a graphics processing unit (GPU) server, an artificial intelligence (AI) server, etc.

[0078] Figure 3 This is a schematic diagram of the hardware structure of a computing device (which may be the server 210 described above) provided in an embodiment of this application. Figure 3 As shown, the computing device may include a processor and a memory; the memory is used to store computer instructions, which are loaded and executed by the processor, enabling the computing device to implement the application building method described in the following embodiments.

[0079] It should be noted that the system architecture and application scenarios described in the embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided in the embodiments of this application. As those skilled in the art will know, with the evolution of system architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0080] Figure 4 This is a flowchart illustrating an application building method provided in an embodiment of this application. For example, the application building method provided in this embodiment can be applied to... Figure 2 In the computing device shown.

[0081] like Figure 4 As shown, the application building method provided in this application embodiment may include the following steps:

[0082] S401. In response to the build request for the target application, obtain the first requirement information input by the user in the first phase of the dialogue.

[0083] The first requirement information describes the capabilities that the target application is expected to achieve.

[0084] In this embodiment, when a user needs to build an application, a build request is initiated on the computing device to start the application build process. Then, the computing device obtains initial requirement information for building the target application through a first-stage dialogue with the user.

[0085] In one possible implementation, the computing device can employ a seventh agent (an information-gathering agent) to converse with the user and obtain initial required information. Specifically, the computing device can display an interactive interface on a display device, and the seventh agent can output dialogue on the interactive interface to guide the user to input the initial required information.

[0086] Specifically, the computing device can use a seventh agent to interact with the user and obtain the user's input; the seventh agent can then determine whether the input meets a third preset condition. If it does, the input is identified as the first required information; if it does not, the seventh agent can prompt the user to re-enter the information. The third preset condition includes: the input is associated with the application and the input clearly and completely describes the capabilities of the target application.

[0087] In other words, the seventh agent can have multiple conversations with the user. After each conversation, it determines whether the user's initial input is related to the application and whether all the input is complete. If all conditions are met, it executes step S402. If not, it continues to have conversations with the user.

[0088] The above implementation method communicates with users in the form of dialogue, thereby gaining a more accurate understanding of user needs and building applications that better meet user requirements.

[0089] In some implementations, the computing device also stores the initial requirement information in a global variable, which is accessible to any intelligent agent. Subsequent intelligent agents can then retrieve this initial requirement information from the global variable for reference, improving the accuracy of application construction.

[0090] Figure 5 This is a schematic diagram illustrating the composition of an information-gathering intelligent agent provided in an embodiment of this application. (Combined with...) Figure 5 To explain, the agent's task is to make an initial judgment on the user's input, determining whether it is relevant to application construction. If relevant, it extracts useful information from the dialogue and writes it into global variables, then performs a completeness check based on the collected data. If incomplete, it continues interacting with the user; if complete, it enters the application construction phase, i.e., executing S402 as follows. For cases where the user's input is irrelevant to application construction, the exception handling function is invoked. This function uses a built-in exception handling function to handle situations where the input is unrelated to application construction. For example, the exception handling function can refuse to answer and guide the user to ask questions relevant to the application construction (e.g., answer: "The content of this input is irrelevant to application construction; please re-enter"). Alternatively, the exception handling function can also be integrated with a chatbot to answer some open-ended questions from the user.

[0091] S402. Based on the first requirement information, generate candidate workflows and display the candidate workflows through a display device.

[0092] Candidate workflows are used to identify the workflow of the target application.

[0093] In this embodiment, after obtaining the first requirement information, the computing device can employ multiple intelligent agents to generate candidate workflows based on the first requirement information. Furthermore, the rendering information of the candidate workflows is sent to a display device, which renders and displays them on the screen to the user, allowing the user to understand the workflow of the target application through the candidate workflows.

[0094] For example, the computing device may employ an agent to select a template from a preset set of workflow templates that matches the first requirement information as a candidate workflow. Alternatively, the computing device may employ another agent to analyze the execution logic in the first requirement information, summarize it, and generate a candidate workflow.

[0095] The specific process of generating candidate workflows using multiple agents is described below and will not be elaborated upon here.

[0096] The workflow is explained in detail below.

[0097] Workflow, also known as a diagram, represents an application in a single diagram, such as... Figure 6 As shown, a workflow consists of nodes and edges. Nodes are responsible for executing specific tasks, while edges represent the flow of those tasks. Users can determine the overall workflow of the application, including what nodes and edges are needed and the order in which they are connected.

[0098] It's important to note that while each node corresponds to a specific task, some tasks have relatively fixed functionalities and cannot be implemented using a single functional node. For example, API calls not only require code support to implement the call but also require the user to manually pass parameters during the call process. In other words, although API calls involve multiple steps, these steps are generally a fixed combination. Therefore, multiple steps can be merged into one step and displayed as a node in the workflow to facilitate workflow construction.

[0099] It's important to note that each node in the application's workflow includes two levels of attributes: a functional description and a functional implementation plan. For example, if the functional description of a node is to translate input Chinese into English, the corresponding implementation plan would be a series of codes that import the input Chinese into the translator and obtain the translation result.

[0100] For a single node, its functionality can be implemented in the following ways: for example, a combination of prompt words and a large model; it can also retrieve results from a specified database using RAG technology; or it can utilize third-party tools (such as translation tools provided by other companies). Therefore, a specific implementation of a node can be derived from one or more combinations of prompt words, large models, codes, RAG, and tools.

[0101] Furthermore, nodes in a workflow can be categorized into functional nodes that include decision-making capabilities and functional nodes that do not. Functional nodes that include decision-making capabilities can be called logical nodes (decision nodes). Logical nodes are used to establish logical connections between functional nodes; for example, based on the output of an upstream functional node, they determine which downstream functional node the task should be routed to.

[0102] Workflow consists of nodes and edges, which are stored in computing devices using different data structures.

[0103] As an example, the data structure of a node is shown in Table 1 below:

[0104] Table 1

[0105]

[0106] As an example, the data structure of edges in the workflow is shown in Table 2 below:

[0107] Table 2

[0108]

[0109] S403. Obtain the second requirement information input by the user in the second stage dialogue, and determine the target workflow based on the second requirement information.

[0110] The second requirement information describes the direction of improvement or confirmation instructions for the candidate workflow, and the target workflow conforms to the second requirement information.

[0111] It should be understood that the initially generated candidate workflow may or may not directly meet the user's needs. Therefore, if it meets the user's needs, the user can confirm it on the display device, and the computing device will then use the candidate workflow as the final target workflow and execute S404 as described below.

[0112] If the user's needs are not met, the user can input directions for improvement on the display device, and the computing device can then further adjust the candidate workflow based on the user's feedback. Furthermore, to facilitate workflow adjustments, this embodiment employs a process adjustment agent to interact with the user, obtain second requirement information, and derive the target workflow based on this second requirement information. For the user, workflow adjustments can be made simply through natural language expression, thereby reducing the professional requirements on the user.

[0113] For example, the process adjustment agent can obtain candidate workflows and second requirement information input by the user, and adjust the candidate workflows according to the second requirement information to generate a new workflow and display it on the display device. Simultaneously, the process adjustment agent will ask the user whether the new workflow is suitable. If suitable, it will execute S404 below; if unsuitable, it will further obtain the user's second requirement information and adjust it again. In other words, the workflow adjustment process can be repeated multiple times; that is, the second-stage dialogue can include multiple dialogues between the agent and the user until the generated workflow meets the user's needs.

[0114] For example, combining Figure 7 Let's explain. If a user wants to develop a Chinese-to-English translation application, the candidate workflow generated by the computing device is: obtain the input Chinese text - translate the Chinese into English - output the English text. The user's second requirement is "adjust to voice output". Based on this second requirement, the process adjustment agent adjusts the candidate workflow, resulting in the new workflow: obtain the input Chinese text - translate the Chinese into English - output the English text - voice broadcast of the English text.

[0115] Optionally, after the process adjustment agent generates a new workflow, it can autonomously determine whether the new workflow meets the user's needs. Alternatively, an additional agent (such as a requirements validation agent) can be set up to determine whether the new workflow meets the user's needs. If it does not, the user is prompted to provide additional information, and then the process adjustment agent generates a new workflow again.

[0116] Optionally, after the process adjustment agent generates a new workflow, it can configure the node agent to determine whether the functional implementation scheme of the newly added nodes in the new workflow can be found in the node template set (which includes multiple node templates, with each node containing a functional implementation scheme corresponding to a specific function). If a scheme can be found, the node template in the set is used. If a scheme cannot be found, an additional configuration agent is introduced to generate the functional implementation scheme for the newly added nodes.

[0117] Specifically, the configuration generation agent can generate a functional implementation scheme (e.g., code implementation) based on the functional description of the newly added node using a large model, and then perform functional verification on the generated functional implementation scheme (or a functional verification agent can be used for verification) to ensure that the functional implementation scheme can run normally. After confirming that it can run normally, the configuration generation agent configures the newly added node.

[0118] For example, Figure 8 This is a schematic flowchart illustrating a process adjustment provided in an embodiment of this application. Figure 8 As shown, the process adjustment agent obtains the second requirement information, modifies the nodes, and outputs a new workflow. The requirement validation agent obtains the new workflow, considers it, and determines whether the new workflow meets the user's requirements. If not, it outputs modification suggestions. The node configuration agent determines whether the new workflow can be found through the node template set. If it cannot be found, it informs the configuration generation agent to generate a functional implementation plan based on the new workflow and performs functional validation.

[0119] S404. Generate the target application based on the target workflow.

[0120] The target workflow meets the second requirement information.

[0121] In this embodiment of the application, the computing device may employ an application-generated intelligent agent to generate a target application based on a target workflow.

[0122] In one possible implementation, the application generation agent prompts the user to input the attribute information required to generate the application. This attribute information includes at least one of the following: name, version number, icon, and online status. Then, based on the target workflow and attribute information, the application generation agent generates the target application, completing the application building process. Additionally, the application generation agent generates an application identifier for the target application according to preset rules. This application identifier can be uniquely identified after the application is launched on the application platform.

[0123] Optionally, the application-generating agent stores the application's attribute information in the application information table of the database to facilitate unified management of applications in the application platform.

[0124] Optionally, when the user inputs attribute information, the application generates an intelligent agent to determine whether the attribute information is missing. If it is missing, the user is prompted to supplement it.

[0125] In one possible implementation, before S404, the method further includes: S405, adjusting the target workflow using a sixth agent to determine whether the target workflow is running normally. When S405 is executed, the aforementioned S404 can be specifically implemented as follows: if it is determined that the target workflow is running normally, generating the target application based on the target workflow.

[0126] Specifically, the above S405 can be implemented as follows:

[0127] S4051. Obtain the initial input information for the target workflow.

[0128] S4052. A sixth intelligent agent is used to output the debugging results of each node sequentially based on the initial input information and the target workflow. The trigger condition for outputting the debugging results of each node is: the process debugging intelligent agent receives a debugging instruction from the user, which instructs the target workflow to execute from the current node to the next node.

[0129] S4053. Based on the adjustment results of each node, determine whether the target workflow is running normally.

[0130] In other words, the process debugging agent can start from the beginning of the target workflow, input initial information, and when the user clicks "Next" on the front-end display device, the process debugging agent can execute the target workflow from the current node to the next node, output the debugging result of the next node, and display it to the user through the display device. If the debugging result of each node can be output and meets expectations (e.g., not empty value, or output error), it means that the target workflow can run normally.

[0131] Figure 9 This is a schematic diagram of a debugging process provided in an embodiment of this application, combined with... Figure 9 Explanation:

[0132] Receive the target workflow and initial input information.

[0133] Input the initial information into the target workflow;

[0134] The target workflow executes the first step.

[0135] Determine whether the current step has reached the end of the target workflow.

[0136] Yes: If the end is reached, the process ends.

[0137] No: If the end has not been reached, check if the user clicks "Next".

[0138] Yes: If the user clicks "Next", then increment the step by one, and the target workflow continues to execute one more step.

[0139] No: If "Next" is not clicked, the user will be asked to take action at this step.

[0140] Repeat the above process until the target workflow is completed, at which point the process ends.

[0141] The process of generating candidate workflows in S402 described above will be explained below with reference to specific embodiments and accompanying drawings.

[0142] In one possible implementation, S402 can be specifically implemented as follows: S4021, the first intelligent agent determines a workflow template that matches the first requirement information from a workflow template set; the workflow template set includes multiple workflow templates corresponding to different capabilities. S4022, a candidate workflow is determined from the candidate workflow templates that match the first requirement information.

[0143] It should be noted that the workflow template set contains multiple preset workflow templates, and each workflow template has corresponding functional description information. For the template matching agent, the task is to match the user's initial requirement information with the existing templates in the set, which is essentially a clustering task. That is, based on the initial requirement information and the functional description information corresponding to each template, the template matching agent performs semantic classification and recall to determine semantically similar candidate workflow templates, and from this, identifies the candidate workflow.

[0144] In one implementation, S4022 above can be specifically implemented as follows: Step a, when there are multiple candidate workflow templates that match the first requirement information, a second agent (template evaluation agent) is used to obtain the third requirement information input by the user in the third stage dialogue. The third requirement information is used to indicate one or more steps included in the expected target application.

[0145] Step b: Using a second intelligent agent, determine the score of each candidate workflow template based on the steps in each candidate workflow template and the third requirement information (few samples); the score is used to characterize the degree of conformity between the candidate workflow template and the third requirement information.

[0146] For example, the second intelligent agent determines the score of each workflow template based on the amount of content including third requirement information in each workflow template; that is, the higher the proportion of content included, the higher the score.

[0147] Step c: Select the candidate workflow templates that meet the first preset criteria as candidate workflows.

[0148] For example, the first preset condition is that the candidate workflow template with the highest score and greater than a certain threshold is used as the candidate workflow.

[0149] Figure 10 This is a schematic diagram illustrating a template matching process provided in an embodiment of this application. For example... Figure 10 As shown, initially, the template matching agent identifies multiple matching templates (e.g., templates 1-4) from the workflow template set. Then, the template evaluation agent scores each template, and the template matching agent determines whether each template meets the criteria (the template with the highest score, whose score is greater than a certain threshold). If the criteria are met, the template with the highest score is output as a candidate workflow. If the criteria are not met, the process ends.

[0150] In one possible implementation, if no candidate workflow template matching the first requirement information exists in the workflow template set, the computing device further performs the following steps: Step 1 Step 2:

[0151] Step 1: Use a third intelligent agent (workflow generation intelligent agent) to obtain the initial workflow based on the first requirement information.

[0152] Step 2: Use the fourth agent (node ​​configuration agent) to configure the functional implementation scheme of each node in the initial workflow to obtain the candidate workflow.

[0153] As mentioned earlier, the workflow used to build the application includes two levels of attributes for each node: a functional description and a functional implementation plan. In step 1 above, the workflow generating agent constructs an initial workflow based on the first requirement information. This initial workflow only provides a workflow at the functional description level; that is, each node in the initial workflow only contains the functional description attributes but not the specific functional implementation plan. Therefore, the application cannot be built based solely on the initial workflow. Therefore, the computing device also performs step 2 above to configure the functional implementation plan for each node in the initial workflow.

[0154] In one implementation, step 2 above can be specifically implemented as follows: for each node in the initial workflow, a fourth agent determines a node template from the node template set that matches the first requirement information, and configures the functional implementation scheme of the node using the node template. The node template set includes multiple node templates, and each node template includes a functional implementation scheme for one function.

[0155] It should be noted that after processing by the workflow generation agent, the functional description, inputs, and outputs of each node in the initial workflow have been defined (related to the first requirement information). The node template set contains multiple preset node templates, each with corresponding functional description information. For the node configuration agent, its task is to match the functional description of each node in the initial workflow with the existing templates in the set; essentially, it's a clustering task. That is, for the target node, the node configuration agent performs semantic classification and recall based on the target node's description and the functional description information corresponding to each template, identifying semantically similar node templates and configuring the functional implementation scheme of those templates onto the target node. This process is repeated until each node of the initial workflow is configured, resulting in a candidate workflow.

[0156] Optionally, it is also necessary to consider whether the functional implementation scheme of the nodes after searching and configuration can operate normally. Therefore, in this embodiment, an additional intelligent agent (such as a functional verification intelligent agent) can be used to verify each node.

[0157] For example, as described above, each node in the initial workflow has defined inputs and outputs. The functional verification agent can generate test information based on the inputs and outputs of each node, input the test information into the node, and check whether the output meets expectations. If it does, it means that the functional implementation scheme of that node can operate normally. If the output does not meet expectations, or if an error occurs, the functional verification agent can instruct the node configuration agent to re-search for the node template, or inform the user of the error so that the user can assist in making modifications.

[0158] Figure 11 This is a schematic diagram illustrating a node configuration process provided in an embodiment of this application. Figure 11 As shown, the process begins with each node in the initial workflow. The node configuration agent determines and configures the node template, and then the function verification agent determines whether the node is running normally. If it can run normally, the process ends; otherwise, the node configuration agent is instructed to search for the node template again.

[0159] Optionally, for the workflow generation agent, some nodes in the generated workflow do not belong to the business layer (i.e., they do not need to be displayed on the front end). For example, combined with Figure 12 The workflow shown is explained. For example... Figure 12As shown, after confirming the validity of the input data, the next step is to preprocess the data (e.g., fill in blank data, remove irrelevant data, etc.), and then select an artificial intelligence (AI) model. The suitability of the model is assessed; if unsuitable, another model is chosen. If suitable, the AI ​​model is trained, and the data is input into the AI ​​model to obtain a data analysis report. However, for users, the more important aspect is the result of data processing—what content will be obtained after data processing. As for how to select an AI model and how to obtain a data analysis report (…),… Figure 12 The content within the dashed box represents functional implementation details, which are not of concern to the user at this stage and therefore do not need to be displayed on the front end. Therefore, to address this situation, this embodiment employs an additional intelligent agent to determine whether each node in the generated workflow needs to be displayed on the front end (whether it belongs to the business logic).

[0160] The computing device also performs the following steps: Step a, using the fifth intelligent agent (business judgment intelligent agent) to determine the target node in the candidate workflow that meets the second preset condition; the second preset condition is used to characterize that the target node does not need to be displayed through the display device.

[0161] Step b: Delete the target node, or integrate the target node with upstream and / or downstream nodes.

[0162] It is understandable that, for target nodes that do not require display on a display device, the fifth agent can remove them from the workflow. For example... Figure 12 For the data preprocessing step, removing this step from the workflow will not affect the user's understanding of the overall workflow. Alternatively, the business decision-making agent can integrate nodes that do not need to be displayed with upstream and downstream nodes, such as... Figure 12 The nodes within the dashed box are considered as a single node (e.g., described as analyzing data).

[0163] In addition, for workflows with a large number of nodes, in order to improve efficiency, multiple business judgment nodes can be set up to work together to determine whether each node in the workflow needs to be displayed through a display device.

[0164] Figure 13 This is a schematic diagram illustrating the operation of a business judgment intelligent agent provided in an embodiment of this application. Figure 13 As shown, the workflow generation agent generates each node based on the first description information, and then integrates each node to output the initial workflow. The business judgment agent (including business judgment agent 1 and business judgment agent 2) interacts with the workflow generation agent, considers the generated nodes during the workflow generation process, determines whether the nodes need to be displayed through a display device, and outputs modification suggestions (deleting nodes or integrating upstream and downstream nodes).

[0165] Figure 14 This is a schematic diagram of a workflow generation method provided in an embodiment of this application. Figure 14 As can be seen, there are two ways to generate workflows. One is to use a template matching agent to search for and determine a matching workflow template from a set of workflow templates. The other is to use a workflow generation agent to generate the workflow, which requires determining node templates from a set of node templates to configure the functional implementation details of the nodes in the generated workflow. Each node template in the set of node templates can be one or more combinations of prompts, large models, codes, RAGs, and tools.

[0166] It should be noted that different intelligent agents are used in the application construction process of this application embodiment, and different intelligent agents play a role in different stages of application construction. Each intelligent agent is obtained by training a large model using specific prompt words, and is essentially a workflow composed of some nodes and edges. As an example, the intelligent agent has the data structure shown in Table 3 below.

[0167] Table 3

[0168]

[0169] The following example illustrates the application construction method of this application embodiment.

[0170] like Figure 15a As shown, on the current interface, guided by the agent, the user can enter the application's name and the capabilities they expect the application to implement. Then, the agent generates a candidate workflow as follows: Figure 15b As shown in the image, the left side displays candidate workflows, and the right side displays controls for debugging workflows. When the user clicks a control, as shown... Figure 15c As shown, users can input the text to be translated, and then the AI ​​agent will determine whether the input is compliant. If it is compliant, it will further search for translation results on the internet and obtain translation results from a large model. The entire adjustment process can be completed within [the specified timeframe]. Figure 15c The interface on the right is designed to help users understand the testing process.

[0171] Figure 16 This is a schematic diagram illustrating a complete application construction process provided in an embodiment of this application. Figure 16As shown, initially, the information gathering agent acquires the first requirement information, and then the template matching agent determines whether a suitable candidate workflow template exists. If a suitable candidate workflow template exists, the process adjustment agent adjusts the process to obtain the target workflow, and then the application generation agent generates the target application. If no suitable candidate workflow template exists, the workflow generation agent generates an initial workflow based on the first requirement information, and then the node configuration agent configures the functional implementation scheme of each node in the initial workflow. Finally, the process adjustment agent adjusts the process to obtain the target workflow, and then the application generation agent generates the target application.

[0172] This application provides an application building method. When a user needs to build a target application, the method first obtains the user's first requirement information from a first-stage dialogue, and then generates candidate workflows based on the first requirement information. Next, it obtains the user's second requirement information from a second-stage dialogue, and then improves or confirms the candidate workflows based on the second requirement information to obtain a target workflow that meets the user's requirements. Finally, it generates the target application based on the target workflow. It can be seen that in the application building process, the solution of this application first uses a first-stage dialogue to comprehensively and accurately obtain the capabilities that the user expects the target application to achieve, thus building an initial workflow. Furthermore, through a second-stage dialogue, it comprehensively and accurately obtains the user's suggestions for improving the initial workflow, and then further adjusts the initial workflow based on the user's suggestions to obtain a target application that more accurately reflects the user's requirements.

[0173] Furthermore, the solution in this application's embodiments addresses the issue of insufficient user guidance by having intelligent agents engage in dialogue with users during the acquisition of application building-related information, thus guiding users to provide complete information. Additionally, during the application building process, a thinking framework is constructed using multiple intelligent agents, which cooperate with each other. Each agent is responsible for different stages of application building, ensuring that the strengths of each agent are fully utilized, further improving the accuracy of application building.

[0174] As can be seen, the above mainly describes the solutions provided by the embodiments of this application from a methodological perspective. To achieve the above functions, the embodiments of this application provide corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the modules and algorithm steps of the various examples described in the embodiments disclosed herein, the embodiments of this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0175] In an exemplary embodiment, this application also provides an application building apparatus. This application building apparatus may be the aforementioned computing device or a processor within the computing device. The application building apparatus may include one or more functional modules for implementing the application building method of the above method embodiments.

[0176] This application also provides a computer-readable storage medium. All or part of the processes in the above method embodiments can be executed by computer instructions instructing related hardware; for example, the related hardware can be a processor of a computing device. The program instructions can be stored in the above-mentioned computer-readable storage medium, and when executed, the processes of the above method embodiments can be implemented. The computer-readable storage medium can be memory. The above-mentioned computer-readable storage medium can also be an external storage device, such as a hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. Further, the above-mentioned computer-readable storage medium can include both memory and external storage devices. The above-mentioned computer-readable storage medium is used to store the above-mentioned computer program instructions and other programs and data required for building the above-mentioned application.

[0177] This application also provides a computer program product comprising a computer program that, when run on a computing device, causes the computing device to execute any of the application building methods provided in the above embodiments.

[0178] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, disclosure, and appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple components. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0179] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of this application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and modifications of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and modifications.

[0180] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An application construction method, characterized in that, The method includes: In response to a build request for a target application, first requirement information input by the user during the first phase of the dialogue is obtained; the first requirement information describes the capabilities that the target application is expected to implement. Based on the first requirement information, a candidate workflow is generated and displayed through a display device; the candidate workflow is used to represent the workflow of the target application. Obtain the second requirement information input by the user in the second stage of the dialogue, and determine the target workflow based on the second requirement information; the second requirement information is used to describe the improvement direction or confirmation indication for the candidate workflow; the target workflow conforms to the second requirement information; The target application is generated based on the target workflow.

2. The method according to claim 1, characterized in that, The step of generating candidate workflows based on the first requirement information includes: A first intelligent agent, based on the first requirement information, determines a workflow template that matches the first requirement information from a workflow template set; the workflow template set includes multiple workflow templates corresponding to different capabilities. The candidate workflow is determined from the candidate workflow templates that match the first requirement information.

3. The method according to claim 2, characterized in that, The step of determining the candidate workflow from the candidate workflow templates that match the first requirement information includes: If there are multiple candidate workflow templates that match the first requirement information, a second agent is used to obtain the third requirement information input by the user in the third stage dialogue. The third requirement information is used to indicate one or more steps included in the expected target application. The second intelligent agent determines a score for each candidate workflow template based on the steps in each candidate workflow template and the third requirement information; the score is used to characterize the degree of conformity between the candidate workflow template and the third requirement information. Candidate workflow templates whose scores meet the first preset conditions are identified as candidate workflows.

4. The method according to claim 2 or 3, characterized in that, If no candidate workflow template matching the first requirement information exists in the workflow template set, the method further includes: A third intelligent agent is used to obtain the initial workflow based on the first requirement information; A fourth intelligent agent is used to configure the functional implementation scheme of each node in the initial workflow based on the first requirement information to obtain the candidate workflow.

5. The method according to claim 4, characterized in that, The scheme for configuring the functionality of each node in the initial workflow using a fourth intelligent agent includes: For each node in the initial workflow, a fourth agent determines a node template that matches the first requirement information from the node template set, and uses the node template to configure the functional implementation scheme of the node; the node template set includes multiple node templates, and each node template includes a functional implementation scheme for one function.

6. The method according to claim 4, characterized in that, Before displaying the candidate workflow via a display device, the method further includes: A fifth intelligent agent is used to determine the target nodes in the candidate workflow that meet the second preset condition, the second preset condition being used to characterize that the target node does not need to be displayed through a display device; Delete the target node, or integrate the target node with upstream and / or downstream nodes.

7. The method according to any one of claims 1-6, characterized in that, The method further includes: Obtain the initial input information of the target workflow; The sixth intelligent agent inputs the initial input information into the target workflow and outputs the debugging results of each node in sequence; wherein, the trigger condition for outputting the debugging results of each node is: the sixth intelligent agent receives the user's debugging instruction, the debugging instruction is used to instruct the target workflow to execute from the current node to the next node; Based on the debugging results of each node, determine whether the target workflow is running normally; The process of generating the target application based on the target workflow includes: If the target workflow is confirmed to be operating normally, the target application is generated based on the target workflow.

8. The method according to any one of claims 1-7, characterized in that, The step of obtaining the first requirement information input by the user in the first stage of the dialogue includes: The seventh intelligent agent is used to communicate with the user and obtain the user's input. A seventh agent is used to determine whether the input content meets a third preset condition. If the condition is met, the input content is determined to be the first requirement information. If the condition is not met, the seventh agent prompts the user to re-enter the information. The third preset condition includes: the input content is associated with the application and the input content can clearly and completely describe the capabilities of the target application.

9. The method according to any one of claims 1-8, characterized in that, The method further includes: The first requirement information is stored in a global variable, which is a variable that any intelligent agent can access.

10. A computing device, characterized in that, The computing device includes a processor and a memory; the processor is coupled to the memory. The memory is used to store computer instructions; The computer instructions are loaded and executed by the processor to enable the computing device to implement the application building method as described in any one of claims 1-9.

Images