Custom workflow architecture and plug-in system based on large language model

Abstract

The invention relates to a custom workflow architecture and plug-in system based on a large language model. The custom workflow architecture and plug-in system comprises a server and an interface interaction display device connected with the server, wherein the server carries a custom workflow architecture module and a plug-in module based on a large language model; the interface interaction display device can interact and display working interfaces of the custom workflow architecture module and the plug-in module; the user-defined workflow architecture module can call plug-ins in the plug-in module; the interface interaction display device can help a user to interact with the custom workflow architecture module in the form of a graphical draggable window, and is used for connecting a plurality of plug-ins in series according to the requirements of the user to form the custom workflow. According to the custom workflow architecture and plug-in system provided by the invention, the graphical drag linker is introduced, so that a user can freely combine and configure various plug-ins to construct personalized workflows.

Description

Technical Field

[0001] The present invention belongs to the field of underground engineering technology, and in particular relates to a custom workflow architecture and plug-in system based on a large language model. Background Art

[0002] With the rapid development of artificial intelligence technology, large language models have been widely used in various scenarios, such as intelligent customer service, information retrieval, and language translation. These technologies can automatically generate text, answer questions, and provide recommendations through natural language processing and machine learning algorithms. However, existing technologies and implementation solutions have some significant problems and limitations. First, existing dialogue systems typically use predefined rules or finite state machines to manage the dialogue process. The main problems with this approach are: 1. Poor flexibility: The dialogue process is predefined and difficult to adapt to dynamically changing user needs. 2. Insufficient scalability: Adding new features or modifying existing features in existing dialogue systems requires a lot of manual adjustments and reconfiguration, which lacks scalability. 3. Poor user experience: Existing dialogue systems have limited understanding of natural language. Users need to ask questions in a specific format, otherwise the system cannot correctly understand and respond.

[0003] Although large language model applications based on existing dialogue systems (such as GPT-4o) perform well in generating natural language text, their application still faces some challenges, mainly manifested in: 1. High dependence on the quality of user input: The quality of the large language model's answers is highly dependent on the way the user asks questions. Different ways of asking questions may cause the model to give completely different answers, which can easily lead to problems of ambiguous information, missing information, and inaccurate information. 2. Lack of custom workflows: Existing large language model applications are usually fixed, and users cannot flexibly customize and combine different functional modules according to their needs. 3. Complex operation: For ordinary users, directly using large language models requires a certain technical background and problem formulation ability, which increases the threshold for use. Summary of the Invention

[0004] (1) Technical issues to be solved

[0005] In response to existing technical problems, the present invention provides a custom workflow architecture and plug-in system based on a large language model.

[0006] (2) Technical solution

[0007] In order to achieve the above objectives, the main technical solutions adopted by the present invention include:

[0008] A custom workflow architecture and plug-in system based on a large language model, comprising: a server and an interface interactive display device connected to the server;

[0009] Among them, the server is equipped with a custom workflow architecture module and plug-in module based on a large language model;

[0010] The interface interactive display device can interact and display the working interface of the custom workflow architecture module and the plug-in module;

[0011] The custom workflow architecture module can call plug-ins in the plug-in module;

[0012] The interface interactive display device can help users interact with the custom workflow architecture module in the form of a graphical draggable window, which is used to connect multiple plug-ins in series to form a custom workflow according to user needs;

[0013] The plug-in module includes: development interface, plug-in library and distribution components;

[0014] The development interface can provide developers with a standard interface for creating new plug-ins;

[0015] The plugin library can store plugins created by developers and packaged in a specific format;

[0016] The distribution component allows developers to package their plug-ins in a specific format and distribute them to other users;

[0017] Among them, the custom workflow architecture module can freely arrange multiple plug-ins in the plug-in library and complete specific process work according to the trigger conditions and working methods set by the user.

[0018] Preferably, the parameters and information of the plug-in on the visual process editing page include: input parameter name, output parameter name and meta information;

[0019] Meta information includes: version number information, author information, and plug-in description information.

[0020] Preferably, the plug-in module further includes: a plug-in loader, a graphical drag-and-drop linker, and a plug-in parameter checker;

[0021] The plug-in loader is responsible for loading the plug-in selected or required by the user from the plug-in library and initializing it for use in the workflow system;

[0022] The graphical drag-and-drop linker provides an intuitive interface that allows users to connect different plug-ins by dragging and dropping to form a complete workflow;

[0023] The plug-in parameter checker is responsible for verifying whether the parameter settings of each plug-in meet the requirements, ensuring that the plug-in can run correctly and work seamlessly with other plug-ins.

[0024] Preferably, the custom workflow architecture module includes: an input processing unit, a natural language understanding unit, a logic processing unit, a natural language generation unit, an output interface unit, and an orchestrator;

[0025] The input processing unit is used to receive and preliminarily process user requests;

[0026] The natural language understanding unit can parse the request using a large language model;

[0027] The logic processing unit can execute the preset business logic according to the analysis results to obtain the processing results;

[0028] The natural language generation unit can convert the processing results into natural language output;

[0029] The output interface unit is used to provide natural language output to the user;

[0030] The orchestrator is used to orchestrate the workflow of multiple plug-ins and set various trigger conditions.

[0031] Preferably, the custom workflow architecture module can identify the input and output of each plug-in and its acceptable data types, and restrict the user to connect only the output type to the output type interface that accepts the corresponding output type.

[0032] Preferably, the plug-in as a whole operates in the form of a Web Service and opens a configuration file interface.

[0033] Preferably, plug-in loading is divided into a pre-loading stage, a loaded stage, and a pre-unloading stage;

[0034] Each stage will call the / preload, / loaded and / unload interfaces in sequence.

[0035] Preferably, the plugin library supports version control and plugin update management;

[0036] The development interface provides support for multiple programming languages, including Python, Java, and JavaScript;

[0037] The input processing unit includes text cleaning and format conversion functions.

[0038] Preferably, the distribution component allows developers to upload plug-ins to the server and download them to other users' devices via the network.

[0039] Preferably, a feedback mechanism is also included to collect user feedback information and optimize system performance and plug-in quality accordingly.

[0040] (3) Beneficial effects

[0041] The beneficial effects of the present invention are:

[0042] 1. System flexibility is significantly improved:

[0043] This application introduces a graphical drag-and-drop connector, allowing users to freely combine and configure various plug-ins to build personalized workflows. Compared to the predefined rules and finite state machine management of traditional dialogue systems, this application's solution greatly improves the system's flexibility, enabling it to adapt to diverse and dynamically changing user needs.

[0044] 2. Enhance system scalability:

[0045] Through a plug-in design and the use of a plug-in loader, this application enables dynamic loading and management of functional modules. Users and developers can easily add new features or optimize existing ones without extensive manual adjustments and reconfiguration. This design significantly improves the system's scalability and enables rapid response to user customization needs.

[0046] 3. Improve user experience:

[0047] This application integrates prompts and guidance functions into the system, providing users with real-time operation guidance and error prompts, helping users understand the functions and parameter settings of each plug-in and avoid configuration errors. In actual application, the prompt function effectively reduces the user's configuration error rate and significantly improves user satisfaction.

[0048] 4. Reduce dependence on the quality of user input:

[0049] The system supports automated data processing and conversion. Users only need to provide raw data, and the system automatically completes the data processing. This reduces the high dependence on the quality of user input, ensures the accuracy and reliability of data processing, and improves the quality of information exchange.

[0050] 5. Lower the user threshold:

[0051] Through the intuitive user interface design, this application significantly reduces the user's learning curve. Even users without a technical background can easily get started. Using prompts, preset templates and automated processing functions, users can quickly build and use customized workflows.

[0052] 6. Improve system reliability and accuracy:

[0053] This application includes a built-in plugin parameter checker to ensure the correct input and output relationships between plugins, reducing operational issues caused by configuration errors. The system's real-time monitoring and feedback mechanism can provide timely information on the operational status of data flows and workflows, helping users quickly identify and resolve issues, thereby improving system stability and reliability.

[0054] 7. The actual application effect is significant:

[0055] In practice, this application's intelligent prompts and automated processing capabilities make user operations simpler and more efficient. Low-code orchestration: Users can build workflows by dragging and dropping plugins, eliminating the need for programming. Type safety: Parameter checkers ensure data type compatibility, avoiding runtime errors. Ecosystem development: Standardized development interfaces and distribution components facilitate plugin sharing and foster a developer community. Intelligent interaction: Integrating large language models enables natural language-driven process processing, lowering the barrier to entry.

[0056] In summary, this application significantly improves the system's flexibility, scalability, user experience, reliability, and accuracy through a series of innovative designs, such as a graphical drag-and-drop linker, plug-in design, prompts and guidance, and automated data processing. Compared to existing technologies, this application provides a more efficient, reliable, and easy-to-use solution that can leverage its advantages in a wide range of application scenarios and meet the diverse needs of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A plug-in diagram of a custom workflow architecture and plug-in system based on a large language model provided by the present invention;

[0058] Figure 2 A schematic diagram of the interface of a plug-in module of a custom workflow architecture and plug-in system based on a large language model provided by the present invention;

[0059] Figure 3 A workflow flowchart of a custom workflow architecture module based on a large language model and a plug-in system provided by the present invention;

[0060] Figure 4 A schematic diagram of an "OR" plug-in of a custom workflow architecture and plug-in system based on a large language model provided by the present invention. DETAILED DESCRIPTION

[0061] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation methods in conjunction with the accompanying drawings.

[0062] like Figures 1-4 As shown: This embodiment discloses a custom workflow architecture and plug-in system based on a large language model, including: a server and an interface interaction display device connected to the server.

[0063] Among them, the server is equipped with a custom workflow architecture module and plug-in module based on a large language model; the interface interaction display device can interact and display the working interface of the custom workflow architecture module and the plug-in module; the custom workflow architecture module can call the plug-in in the plug-in module; the interface interaction display device can help users interact with the custom workflow architecture module in the form of a graphical draggable window, so as to connect multiple plug-ins in series according to user needs to form a custom workflow.

[0064] The plug-in module includes a development interface, a plug-in library, and a distribution component. The development interface provides developers with a standard interface for creating new plug-ins. The plug-in library stores plug-ins created by developers and packaged in a specific format. The distribution component allows developers to package their plug-ins in a specific format and distribute them to other users. The custom workflow architecture module allows for the free arrangement of multiple plug-ins in the plug-in library and completes specific workflows based on user-defined trigger conditions and working methods.

[0065] Through the system in this embodiment, developers and users can freely arrange plug-ins, set various trigger conditions, and configure plug-in operation modes to achieve the goal of mixing multiple plug-ins to complete specific process tasks. Through the plug-in system, users can create their own plug-ins for various needs based on the development interface provided by the system, and can package their own code in a specific format and distribute it to other users. Users use plug-ins and use the system's orchestrator to freely connect them to the large language model for use.

[0066] In this embodiment, the parameters and information of the plug-in on the visual process editing page include: input parameter name, output parameter name and meta information. The meta information includes: version number information, author information, and plug-in description information.

[0067] The input and output sections include the accepted types and parameter names of the input or output parameters. Accepted types depend on the types accepted by the large language model. For example, if the large language model can process text, images, audio, video, and other file types, the plug-in system's response data types may also accept text, images, audio, video, and other file types. Parameter names are displayed in the subsequent workflow system to facilitate programmers to distinguish the purpose of the interface by name.

[0068] like Figure 2 As shown: Taking a plug-in that converts video to text as an example, the input type is video, the output type is text, the input and output parameters are customizable, and the original information includes the version number, author, and plug-in description, and the content can be customized.

[0069] In this embodiment, the plug-in module also includes: a plug-in loader, a graphical drag-and-drop linker, and a plug-in parameter checker; the plug-in loader is responsible for loading the plug-in selected or required by the user from the plug-in library and initializing it for use in the workflow system; the graphical drag-and-drop linker can provide an intuitive interface, allowing users to connect different plug-ins through drag-and-drop operations to form a complete workflow; the plug-in parameter checker is responsible for verifying whether the parameter settings of each plug-in meet the requirements, ensuring that the plug-in can run correctly and work seamlessly with other plug-ins.

[0070] The custom workflow architecture module in this embodiment includes: an input processing unit, a natural language understanding unit, a logic processing unit, a natural language generation unit, an output interface unit, and an orchestrator;

[0071] Among them, the input processing unit is used to receive and preliminarily process user requests; the natural language understanding unit can use a large language model to parse the request; the logic processing unit can execute the preset business logic according to the parsing result to obtain the processing result; the natural language generation unit can convert the processing result into natural language output; the output interface unit is used to provide natural language output to the user; the orchestrator is used to orchestrate the workflow of multiple plug-ins and set various trigger conditions.

[0072] In a custom workflow architecture and plugin system based on a large language model, the plugin loader, the graphical drag-and-drop linker, and the plugin parameter inspector are key components. The connection and coordination between them are crucial to the system's flexibility, ease of use, and reliability. The following is a detailed description of how these three components interact:

[0073] 1. Plugin Loader

[0074] Function: The plug-in loader is responsible for loading the plug-in selected or required by the user from the plug-in library and initializing it for use in the workflow system.

[0075] Interaction:

[0076] Interaction with the plug-in library: Get the required plug-in package from the plug-in library (possibly downloaded from the network or read locally).

[0077] Interaction with the orchestrator: Provide the loaded plug-in to the orchestrator so that users can orchestrate and manage the plug-in through the orchestrator.

[0078] Interaction with the plug-in parameter inspector: When loading a plug-in, the plug-in parameter inspector is called to verify that the plug-in configuration is correct.

[0079] 2. Graphical drag linker

[0080] Function: The graphical drag-and-drop linker provides an intuitive interface that allows users to connect different plug-ins by dragging and dropping to form a complete workflow.

[0081] Interaction:

[0082] Interaction with the orchestrator: The orchestrator updates the order and connection relationships of the plug-ins in the workflow based on the user's drag-and-drop operations.

[0083] Interaction with the plug-in loader: When the user selects a plug-in from the plug-in library and drags it to the workspace, the plug-in loader is notified to load the corresponding plug-in.

[0084] Interaction with the plug-in parameter checker: After the user completes the plug-in connection, the plug-in parameter checker is triggered to perform final parameter verification to ensure that all connected plug-ins are configured correctly.

[0085] 3. Plugin Parameter Inspector

[0086] Function: The plug-in parameter checker is responsible for verifying whether the parameter settings of each plug-in meet the requirements, ensuring that the plug-in can run correctly and work seamlessly with other plug-ins.

[0087] Interaction:

[0088] Interaction with plugin loader: During the plugin loading process, the plugin parameter checker is called to verify that the initial configuration of the plugin is correct.

[0089] Interaction with the graphical drag-and-drop linker: When the user completes the connection and configuration of the plug-in, the plug-in parameter checker is called again to verify whether the parameter settings of all plug-ins in the entire workflow are reasonable and compatible.

[0090] Interaction with the orchestrator: Feedback inspection results to the orchestrator. If any problems are found, the orchestrator can prompt the user to make adjustments.

[0091] Cooperation process example

[0092] (1) Plug-in selection and loading:

[0093] The user selects the required plug-in from the plug-in library through a graphical drag-and-drop linker and drags it to the workspace.

[0094] The plugin loader loads these plugins from the plugin library according to the user's selection and calls the plugin parameter checker for preliminary parameter verification.

[0095] (2) Plug-in arrangement:

[0096] Users can connect loaded plug-ins according to business logic by dragging and dropping to form a complete workflow.

[0097] The orchestrator records the connection relationships between plug-ins and dynamically adjusts the workflow structure based on user operations.

[0098] (3) Parameter verification and optimization:

[0099] After the user completes the plug-in connection, the graphical drag linker triggers the plug-in parameter checker for comprehensive parameter verification.

[0100] The Plug-in Parameter Inspector analyzes each plug-in's input and output ports to ensure they are passing data correctly between them, and prompts the user to correct any incompatible configurations.

[0101] (4) Execution and feedback:

[0102] After all configurations and verifications are completed, users can start the workflow and the orchestrator will execute the functions of each plug-in in sequence.

[0103] The system collects feedback information during operation for users to further optimize and adjust workflows.

[0104] It should be noted that the custom workflow architecture module in this embodiment can identify the input and output of each plug-in and the data types it accepts, and restricts users to connecting output types to output type interfaces that accept that type. That is, if the output type of the current plug-in is text, the input type of the next plug-in in the workflow must also be text. This ensures that plug-ins can automatically complete information transmission during specific workflow processes, allowing users to conveniently use the plug-in system without having to understand or understand the workings of the plug-ins.

[0105] If a user wants to understand the content of a video through a large language model and wants the large language model to give a text description of the video, the workflow of this behavior is as follows: Figure 3 shown.

[0106] The system also provides plug-in templates such as AND, OR, and NOT. Figure 4 As shown: Taking the OR plug-in as an example, the plug-in receives the output of two different plug-ins as the input of the OR plug-in. If one of the two plug-ins can output, the entire OR plug-in will output a content, thereby ensuring that the workflow system as a whole has a certain logical capability.

[0107] In this embodiment, the plug-in operates as a Web Service, exposing a configuration file interface. Plug-in loading is divided into preload, loaded, and pre-unload phases; each phase calls the / preload, / loaded, and / unload interfaces in sequence.

[0108] The plug-in library in this embodiment supports version control and plug-in update management; the development interface provides support for multiple programming languages, including Python, Java, and JavaScript; and the input processing unit includes text cleaning and format conversion functions.

[0109] The distribution component allows developers to upload plug-ins to the server and download them to other users' devices via the network.

[0110] The system in this embodiment also includes a feedback mechanism for collecting user feedback information and optimizing system performance and plug-in quality accordingly.

[0111] The system component interaction process in this embodiment is as follows:

[0112] Plugin development and distribution:

[0113] Developers write plug-ins through the development interface, define input and output parameters and data types (for example, the input type of a video-to-text plug-in is "video" and the output is "text"); the plug-in is packaged in a specific format and uploaded to the server, pushed to other users through the distribution component, and the plug-in library records version information.

[0114] Workflow creation:

[0115] The user drags the plug-in to the workspace through the interface interaction device, and the graphical drag linker triggers the plug-in loader to load the corresponding plug-in; the plug-in parameter checker verifies the initial configuration (such as whether the parameter name and data type are legal) during the loading phase; the user connects the plug-in by dragging and dropping, and the orchestrator records the connection relationship (for example, the input type of the next plug-in must match the output type of the previous plug-in).

[0116] Process execution and verification:

[0117] The input processing unit receives user requests (such as "generate a text description of the video"), cleans the text and converts the format; the natural language understanding unit parses the request, and the logic processing unit calls the "video to text" plug-in, combined with the "or" plug-in to process multi-source input; the plug-in parameter checker performs secondary verification after the connection is completed to ensure that the data type matches (such as the video output is correctly connected to the text input); the natural language generation unit converts the processing results into text and returns it to the user through the output interface.

[0118] Plugin lifecycle management:

[0119] In the preloading phase, the / preload interface is called to initialize resources; in the loaded phase, the plug-in function is activated through the / loaded interface; in the pre-unloading phase, the / unload interface is called to release resources and support dynamic expansion.

[0120] The technical principles of the present invention have been described in conjunction with specific embodiments. These descriptions are intended solely to illustrate the principles of the present invention and are not to be construed in any way as limiting the scope of protection of the present invention. Based on the explanations herein, those skilled in the art will be able to devise other specific embodiments of the present invention without inventive effort, and such embodiments will fall within the scope of protection of the present invention.

Claims

1. A custom workflow architecture and plug-in system based on a large language model, characterized by: include: A server and an interface interactive display device connected to the server; Among them, the server is equipped with a custom workflow architecture module and plug-in module based on a large language model; The interface interactive display device can interact and display the working interface of the custom workflow architecture module and the plug-in module; The custom workflow architecture module can call plug-ins in the plug-in module; The interface interactive display device can help users interact with the custom workflow architecture module in the form of a graphical draggable window, which is used to connect multiple plug-ins in series to form a custom workflow according to user needs; The plug-in module includes: development interface, plug-in library and distribution components; The development interface can provide developers with a standard interface for creating new plug-ins; The plugin library can store plugins created by developers and packaged in a specific format; The distribution component allows developers to package their plug-ins in a specific format and distribute them to other users; Among them, the custom workflow architecture module can freely arrange multiple plug-ins in the plug-in library and complete specific process work according to the trigger conditions and working methods set by the user.

2. The custom workflow architecture and plug-in system according to claim 1, characterized in that: The parameters and information of the plug-in on the visual process editing page include: input parameter name, output parameter name and meta information; Meta information includes: version number information, author information, and plug-in description information.

3. The custom workflow architecture and plug-in system according to claim 2, characterized in that: The plug-in module also includes: plug-in loader, graphical drag linker, plug-in parameter checker; The plug-in loader is responsible for loading the plug-in selected or required by the user from the plug-in library and initializing it for use in the workflow system; The graphical drag-and-drop linker provides an intuitive interface that allows users to connect different plug-ins by dragging and dropping to form a complete workflow; The plug-in parameter checker is responsible for verifying whether the parameter settings of each plug-in meet the requirements, ensuring that the plug-in can run correctly and work seamlessly with other plug-ins.

4. The custom workflow architecture and plug-in system according to claim 3, characterized in that: The custom workflow architecture module includes: input processing unit, natural language understanding unit, logic processing unit, natural language generation unit, output interface unit and orchestrator; The input processing unit is used to receive and preliminarily process user requests; The natural language understanding unit can parse the request using a large language model; The logic processing unit can execute the preset business logic according to the analysis results to obtain the processing results; The natural language generation unit can convert the processing results into natural language output; The output interface unit is used to provide natural language output to the user; The orchestrator is used to orchestrate the workflow of multiple plug-ins and set various trigger conditions.

5. The custom workflow architecture and plug-in system according to claim 4, characterized in that: The custom workflow architecture module can identify the input and output of each plug-in and its acceptable data types, and restrict users to connect the output type to the output type interface that accepts the corresponding type.

6. The custom workflow architecture and plug-in system according to claim 5, characterized in that: The plug-in operates as a Web Service and opens a configuration file interface.

7. The custom workflow architecture and plug-in system according to claim 6, characterized in that: Plugin loading is divided into pre-loading stage, loaded stage and pre-unloading stage; Each stage will call the / preload, / loaded and / unload interfaces in sequence.

8. The custom workflow architecture and plug-in system according to claim 1, characterized in that: The plug-in library supports version control and plug-in update management; The development interface provides support for multiple programming languages, including Python, Java, and JavaScript; The input processing unit includes text cleaning and format conversion functions.

9. The custom workflow architecture and plug-in system according to claim 1, characterized in that: The distribution component allows developers to upload plug-ins to the server and download them to other users' devices over the network.

10. The custom workflow architecture and plug-in system according to claim 1, characterized in that: It also includes a feedback mechanism for collecting user feedback and optimizing system performance and plug-in quality accordingly.

Images