AI large model and low-code platform interactive integration method and system

Abstract

The invention provides an interactive integration method and system for an AI large model and a low-code platform, and relates to the technical field of interaction of the AI large model and the low-code platform, and the method comprises the steps: obtaining a task demand input by a user, analyzing the task demand into a structured task instruction, automatically matching an optimal AI model, and dynamically generating an adaptive API call parameter, according to the method, the AI large model and the low-code platform can be deeply integrated, the technical problems of integration complexity, interaction mode solidification, updating lag and the like are solved, and the method has the advantages that the method is simple and convenient to operate, and the method is suitable for popularization and application. According to the method, a user can call a complex model without AI professional knowledge, the low-code development efficiency is improved by more than 50%, and the security in a data processing process is enhanced by introducing a security verification mechanism, so that the integration efficiency of a low-code platform and an AI large model is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of interaction between AI big models and low-code platforms, and in particular to a method and system for interactive integration of AI big models and low-code platforms. Background Art

[0002] With the rapid development of artificial intelligence technology, low-code platforms have become one of the important tools for developers to build applications. Low-code platforms simplify the coding process, allowing non-professional developers to quickly build and customize applications, thereby significantly improving development efficiency and lowering technical barriers. However, existing low-code platforms still face some technical challenges when integrating large AI models.

[0003] For example, existing low-code platforms mainly rely on preset modules and visual logic orchestration, which makes it difficult to directly call complex AI models (such as natural language processing, image generation, etc.). Developers are required to have AI expertise, which limits the intelligent capabilities of low-code platforms. Traditional integration methods (such as API calls) require manual configuration of parameters and data processing procedures, lack dynamic adaptation capabilities, and cannot automatically optimize interaction logic based on user needs or scenario changes; AI large models are updated and iterated quickly, but low-code platforms lack a dynamic synchronization mechanism, which requires redeployment or adjustment of platform logic after model upgrades, resulting in a long development cycle; In the prior art, data interaction between AI big models and low-code platforms may expose sensitive information and lack an end-to-end security isolation mechanism. Therefore, the present invention proposes an interactive integration method and system for AI big models and low-code platforms to solve the problems existing in the prior art. Summary of the invention

[0004] In response to the above problems, the purpose of the present invention is to propose a method and system for the interactive integration of AI big models and low-code platforms. This method for the interactive integration of AI big models and low-code platforms can lower the technical threshold for the integration of AI big models and low-code platforms, and can solve the problems existing in the prior art.

[0005] To achieve the purpose of the present invention, the present invention is implemented by the following technical solution: a method for interactive integration of a large AI model and a low-code platform, comprising the following steps: Step 1: Obtain the task requirements input by the user and parse them into structured task instructions; Step 2: According to the analysis results of step 1, automatically match the optimal AI model and dynamically generate adaptive API call parameters; Step 3: Based on the parameters generated in step 2, map them to standardized components of the low-code platform; Step 4: In a sandbox environment, verify the security and functionality of standardized components, and automatically deploy them to the low-code platform after confirmation.

[0006] A further improvement is that in step 1, the structured task instructions include target functions, input and output formats, and constraints.

[0007] A further improvement is that in step 2, the specific method of automatic matching is: S1. Build a dynamically updated model capability map to record the metadata of all available AI models; S2. Arrange the vector of user task attributes into query conditions and search for matching subgraphs in the model capability graph vector.

[0008] A further improvement is that in step 2, a reinforcement learning mechanism is introduced to optimize the model selection strategy based on historical interaction data.

[0009] A further improvement is that the metadata includes functional descriptions, performance indicators and compliance attributes.

[0010] A system for interactive integration of an AI large model and a low-code platform, including an intelligent interactive middle layer, wherein the intelligent interactive middle layer includes: A natural speech interaction interface module is used to parse the input task requirements into structured task instructions; Dynamic adaptation engine module, which is used to automatically match the optimal AI model based on the analysis results and dynamically generate adaptive API call parameters; A visual logic orchestrator module is used to map the generated parameters into standardized components of the low-code platform; The security verification and deployment module is used to verify the security and functionality of standardized components, and automatically deploy them to the low-code platform after confirmation.

[0011] A further improvement is that the natural speech interaction interface module has a built-in pre-trained language model for intent recognition and semantic disambiguation.

[0012] A further improvement is that the standardized components include data tables, flow charts and UI controls.

[0013] The beneficial effects of the present invention are as follows: the core of the present invention is to build an intelligent interactive middle-layer system to achieve deep integration of AI big models and low-code platforms. Through the combination of modules such as dynamic adaptation engine, natural language interactive interface, and visual logic orchestrator, it not only solves technical problems such as integration complexity, interaction mode solidification, and update lag, but also enables users to call complex models without AI expertise, and improves low-code development efficiency by more than 50%. By introducing a security verification mechanism, the security of the data processing process is enhanced, thereby greatly improving the integration efficiency and reliability of the low-code platform and AI big models. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic diagram of the steps of the present invention.

[0015] Figure 2 It is a schematic diagram of the interaction process of the present invention. DETAILED DESCRIPTION

[0016] In order to deepen the understanding of the present invention, the present invention will be further described in detail below in conjunction with examples. The examples are only used to explain the present invention and do not constitute a limitation on the protection scope of the present invention.

[0017] Embodiment 1, according to Figure 1-Figure 2 As shown, this embodiment proposes a method for interactive integration of an AI large model and a low-code platform, including the following steps: Step 1: Obtain the task requirements input by the user and parse them into structured task instructions. Specifically, obtain the task requirements input by the user through natural language (such as voice or text), such as generating a module that automatically classifies according to user comments, and parse it into structured task instructions, including target functions, input and output formats, and constraints. The target function is the specific function that the user expects to achieve, such as classification tasks, image generation, etc. The input and output formats include the type and structure of the input data and the requirements of the output data (such as prediction results, generated content), and the constraints include specific constraints such as time limits and accuracy requirements.

[0018] Step 2: According to the analysis results of step 1, automatically match the optimal AI model and dynamically generate adaptive API call parameters. For example, BERT is used for text classification and StableDiffusion is used for image generation. Generate adaptive API call parameters according to the specific requirements of the task. Specifically, the specific method of automatic matching is as follows: S1. Build a dynamically updated model capability map to record the metadata of all available AI models, including functional descriptions (supported inputs / outputs, applicable scenarios), performance indicators (accuracy, latency, resource consumption), and compliance attributes (whether federated learning and differential privacy are supported), to help the system select the appropriate model; S2. Arrange the vector of user task attributes (task type, task input, data modality) into query conditions and search for matches in the model capability graph vector. For example, the task is "image style transfer" → the matching subgraph contains nodes such as StableDiffusion.

[0019] A reinforcement learning mechanism is introduced to optimize the model selection strategy based on historical interaction data to improve matching accuracy and efficiency. Specifically, the Bayesian optimization algorithm is combined to dynamically adjust the model calling parameters according to the historical task execution results.

[0020] The reinforcement learning mechanism is used in this scenario to learn and optimize decision-making strategies through interaction with the environment. The system uses historical interaction data (such as task type, input data, selected model and its performance, etc.) as training data to update the model selection strategy. Specifically, the present invention uses the execution effect of historical tasks (including success rate, execution time, model accuracy, etc.) as a "reward signal" to input into the reinforcement learning model. The goal of the reinforcement learning model is to select the optimal AI model in future tasks by optimizing the "reward signal".

[0021] Bayesian optimization is a global optimization algorithm based on Bayesian theory, which is suitable for high-dimensional, non-convex, and computationally expensive optimization problems. In this scenario, the Bayesian optimization algorithm is mainly used to dynamically adjust the calling parameters of the AI ​​model to improve the accuracy and efficiency of task execution.

[0022] For example, when a user submits a new task whose goal is to perform sentiment analysis on a text, the system evaluates and selects the model that best suits the sentiment analysis task (for example, the BERT model) based on the performance of historical tasks. When executing the task, the system uses the Bayesian optimization algorithm to dynamically adjust the parameters of the BERT model (such as learning rate, batch size, etc.) to achieve optimal performance (such as improving accuracy or reducing computing resource consumption). After the model executes the task and outputs the results, the system feeds back the execution effect of the task (such as accuracy, latency, etc.) to the reinforcement learning and Bayesian optimization models, updates the parameters and strategies, and gradually improves the accuracy and efficiency of task execution.

[0023] Step 3: Based on the parameters generated in step 2, map them into standardized components of the low-code platform, including data tables, flowcharts, and UI controls, where: Data table: convert the classification results into a structured data table for further analysis; Flowchart: Convert the content generated by the model (such as decision nodes, classification branches, etc.) into a flowchart for users to further operate and customize; UI controls: present the generated content as UI elements to facilitate user interaction.

[0024] The further dynamic adaptation engine module provides a visual operation interface, which allows users to fine-tune these standardized components and feed them back to the AI ​​model to further optimize task results.

[0025] Step 4: After task execution and result conversion, all standardized components enter the sandbox environment for security and functionality verification. This verification process ensures that the components do not cause any security issues (such as data leakage or malicious attacks) and can run efficiently in actual applications. After verification, the system will automatically deploy these components to the low-code platform. Users can further customize, debug and use them through the low-code platform.

[0026] Furthermore, in order to ensure the security of user data and privacy, encryption measures will be taken for data interaction and task execution throughout the entire process. At the same time, sandbox verification will be carried out in an isolated environment to avoid any potential risks.

[0027] Embodiment 2: This embodiment proposes a system for interactive integration of a large AI model and a low-code platform. The system includes an intelligent interactive middle layer, which includes a natural speech interactive interface module for parsing into structured task instructions, a dynamic adaptation engine module for matching the optimal AI model, and a visual logic orchestrator module for converting standard components of the low-code platform, wherein the standardized components include data tables, flow charts, and UI controls. A pre-trained language model (such as GPT-4) for intent recognition and semantic disambiguation is built into the natural speech interactive interface module.

[0028] Specific: Natural Language Interaction Interface (NLUI): used to parse the user's natural language input and convert it into structured instructions that the machine can understand; Dynamic Adaptation Engine Module (DAE): automatically matches the appropriate AI model according to user needs and generates corresponding API call parameters; Visual Logic Orchestrator Module (VLE): Converts the output results of the AI ​​model into standardized components in the low-code platform and provides a visual operation interface to facilitate users to fine-tune and optimize.

[0029] Security verification and deployment module: used to verify the security and functionality of standardized components, and automatically deploy them to the low-code platform after confirmation.

[0030] As AI models are iteratively updated, low-code platforms can automatically identify and adapt to the latest model versions, reduce manual intervention, and improve work efficiency. Through natural language interfaces and visual orchestration tools, the system allows users to perform complex tasks through simple operations without having to deeply understand the technical details of AI models. The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the framework and scope of application of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for interactive integration of a large AI model and a low-code platform, characterized by: The following steps are involved: Step 1: Obtain the task requirements input by the user and parse them into structured task instructions; Step 2: According to the analysis results of step 1, automatically match the optimal AI model and dynamically generate adaptive API call parameters; Step 3: Based on the parameters generated in step 2, map them to standardized components of the low-code platform; Step 4: In a sandbox environment, verify the security and functionality of standardized components, and automatically deploy them to the low-code platform after confirmation.

2. According to claim 1, a method for interactive integration of a large AI model and a low-code platform is characterized in that: In the step 1, the structured task instructions include target functions, input and output formats, and constraints.

3. According to claim 1, the method for interactive integration of a large AI model and a low-code platform is characterized in that: In step 2, the specific method of automatic matching is: S1. Build a dynamically updated model capability map to record the metadata of all available AI models; S2. Arrange the vector of user task attributes into query conditions and search for matching subgraphs in the model capability graph vector.

4. According to claim 1, a method for interactive integration of a large AI model and a low-code platform is characterized in that: In the step 2, a reinforcement learning mechanism is introduced to optimize the model selection strategy based on historical interaction data.

5. According to claim 3, the method for interactive integration of a large AI model and a low-code platform is characterized in that: The metadata includes functional descriptions, performance indicators, and compliance attributes.

6. A system for interactive integration of AI big model and low-code platform, characterized by: The intelligent interaction middle layer includes: A natural speech interaction interface module is used to parse the input task requirements into structured task instructions; Dynamic adaptation engine module, which is used to automatically match the optimal AI model based on the analysis results and dynamically generate adaptive API call parameters; A visual logic orchestrator module is used to map the generated parameters to standardized components of the low-code platform; The security verification and deployment module is used to verify the security and functionality of standardized components, and automatically deploy them to the low-code platform after confirmation.

7. The system of the interactive integration method of a large AI model and a low-code platform according to claim 6 is characterized by: The natural speech interaction interface module has a built-in pre-trained language model for intent recognition and semantic disambiguation.

8. The system of the interactive integration method of a large AI model and a low-code platform according to claim 6 is characterized by: The standardized components include data tables, flow charts, and UI controls.

Images