AI capability calling method and system based on ai model intelligent review management and control platform, and storage medium

Abstract

The application discloses an AI capability calling method and system based on an AI model intelligent review management and control platform and a storage medium, wherein the method acquires configurable condition parameters from the platform through a business system; then, specific task parameters and a result callback address are submitted. The platform intelligently matches preset rules according to the parameters, dynamically determines AI models and prompt words to be called, and places the task into an asynchronous queue for scheduling, so as to effectively control the load and prevent congestion. The platform can concurrently call multiple AI models, integrates and processes the returned results, and finally returns unified results to the business system through the callback address. Meanwhile, the platform records full-link logs, supports effect monitoring and model optimization. The application realizes unified access, flexible configuration, intelligent scheduling and centralized management and control of AI capabilities by decoupling the business and the model, and significantly improves the integration efficiency, system stability and operation management visibility.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, system and storage medium for invoking AI capabilities based on an AI model-based intelligent audit and management platform. Background Technology

[0002] With the widespread application of artificial intelligence technology across various industries, it has become commonplace for enterprises to integrate multiple AI models (including general-purpose large models and vertical-domain small models) into their various business systems to achieve intelligentization. Currently, the mainstream integration methods and their existing technical shortcomings are as follows:

[0003] Patent CN113050955A discloses a method for dynamically selecting the AI ​​model executor and deployment location on the edge, making adaptive decisions for single calls based on local resource status. However, this solution is a distributed strategy embedded in specific business terminals, rather than a centralized platform providing unified service access, centralized scheduling, and management for multiple backend business systems. Therefore, it cannot decouple business systems from AI models, nor can it perform cross-business resource coordination and data analysis.

[0004] Patent CN117649538A discloses a method and apparatus for matching industrial quality inspection scenarios with AI models, focusing on feature matching and recommendation in the model selection stage. However, this solution is essentially a "model selection system" and does not address how the selected model can be efficiently, stably, and controllably invoked by the business system. It lacks design for key runtime aspects such as asynchronous scheduling, queue management, result integration, and end-to-end monitoring in the invocation process.

[0005] In summary, existing technologies lack a centralized management platform solution that can sit between diverse business systems and diverse AI models, enabling unified access, intelligent scheduling, dynamic configuration, and data closure. This results in enterprises still facing systemic challenges when integrating AI capabilities, including deep system coupling, low development and operation efficiency, uneven resource utilization, and long model iteration cycles. Summary of the Invention

[0006] The purpose of this invention is to provide an AI capability invocation method, system, and storage medium based on an AI model-based intelligent audit and management platform. It adopts a microservice architecture to achieve functions such as unified access, dynamic configuration, capability orchestration, rate limiting control, monitoring logs, and data feedback for model invocation.

[0007] The technical solution adopted in this invention is:

[0008] The AI ​​capability invocation method based on the AI ​​model-based intelligent review and control platform is applied to a system including business systems and the AI ​​model-based intelligent review and control platform. The method includes:

[0009] The business system initiates the first request to the AI ​​model-based intelligent review and control platform;

[0010] The AI ​​model intelligent review and control platform returns pre-configured condition parameters for the AI ​​task to the business system based on the first request. The condition parameters include filterable conditions and prompt word template parameters.

[0011] The business system determines the specific request parameter values ​​based on the condition parameters and sends a second request to the AI ​​model intelligent audit and control platform. The second request carries the request parameter values ​​and the result callback address.

[0012] The AI ​​model intelligent review and control platform matches the request parameter values ​​with a pre-set rule base to determine at least one target AI model and its corresponding prompt words, and adds the AI ​​task to the asynchronous task queue.

[0013] The AI ​​model intelligent review and control platform calls the AI ​​model base and asynchronously executes AI tasks in the task queue to obtain results from at least one target AI model.

[0014] The AI ​​model-based intelligent review and control platform processes the results and returns them to the business system via a result callback address.

[0015] Furthermore, the AI ​​model intelligent review and control platform matches the request parameter values ​​with a pre-set rule base to determine at least one target AI model and its corresponding prompt words, including:

[0016] The request parameter values ​​are matched with multiple predefined rules in the rule pool. The rules define the mapping relationship between business scenarios, judgment conditions, AI models, and prompt words.

[0017] Filter out all matching rules, and obtain the corresponding target AI model and prompt word configuration based on the rules, and package them into a task list.

[0018] Furthermore, the rules include a first category of rules that must be executed unconditionally and a second category of rules that are triggered based on content characteristics in the request parameter values.

[0019] Furthermore, the implementation of adding AI tasks to the asynchronous task queue includes:

[0020] Check if the current number of queued tasks has reached the threshold;

[0021] If the threshold is not reached, the encapsulated task list will be placed in the specified task queue for scheduling; if the threshold is reached, an error message will be returned.

[0022] Furthermore, the AI ​​model intelligent review and control platform processes the results by integrating and analyzing multiple results from multiple target AI models when there are multiple target AI models, forming a unified result.

[0023] Furthermore, the intelligent audit and control platform records the entire call log from receiving the second request to returning the result. The log includes the request parameter values, matching rules, the AI ​​model called, the time consumed, the return results of each model, and the final result.

[0024] The AI ​​capability invocation system based on the AI ​​model-based intelligent review and control platform includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the AI ​​capability invocation method based on the AI ​​model-based intelligent review and control platform.

[0025] A computer-readable storage medium storing a computer program, wherein when executed by a processor, the computer program implements the steps of an AI capability invocation method for an AI model-based intelligent audit and control platform.

[0026] The present invention adopts the above technical solution and has the following beneficial effects compared with the prior art:

[0027] 1) The platform centrally manages the access, authentication, monitoring, and alerting of all AI models, eliminating the previous need to configure and maintain multiple model connections separately for each business system, significantly reducing system operation and maintenance complexity and costs. 2) Interaction is achieved through a unified, standard API provided by the platform, completely decoupling from the underlying specific AI models. When adding or replacing AI models or optimizing prompts, no modification to any business system code is required; the entire process can be completed visually through the platform's configuration interface, shortening the development, testing, and deployment cycle from weeks to hours, significantly improving demand response efficiency. 3) Through the dynamic configuration function of rules and prompts provided on the front end, business personnel can quickly respond to policy changes. For newly emerging content risks, new rules can be configured immediately and associated with the corresponding models. 4) The platform provides a queuing function to prevent AI models from experiencing congestion and performance issues during peak periods; it also records complete logs of AI capability calls, providing test cases for subsequent AI optimization; and it provides a large-screen reporting function, fully displaying the effects of AI capability calls.

[0028] This invention employs a microservice architecture to achieve unified access, dynamic configuration, capability orchestration, rate limiting, monitoring logs, and data feedback for model calls. Business systems can invoke various AI capabilities simply through the platform's APIs, without needing to concern themselves with details such as underlying model switching and prompt word adjustments. The platform boasts excellent scalability and high performance, fully meeting the call demands during peak business periods. Attached Figure Description

[0029] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments;

[0030] Figure 1 This is a flowchart illustrating the AI ​​capability invocation method of the AI ​​model-based intelligent audit and control platform of the present invention;

[0031] Figure 2 This is a schematic diagram of the intelligent scheduling rule calculation process;

[0032] Figure 3 This is a schematic diagram of the intelligent queuing and scheduling process. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0034] like Figures 1 to 3 As shown in any of the accompanying drawings, a specific embodiment of the present invention discloses an AI capability invocation method based on an AI model-based intelligent review and control platform, applicable to a system including a business system and an AI model-based intelligent review and control platform. The method includes:

[0035] The business system initiates the first request to the AI ​​model-based intelligent review and control platform;

[0036] The AI ​​model intelligent review and control platform returns pre-configured condition parameters for the AI ​​task to the business system based on the first request. The condition parameters include filterable conditions and / or prompt word template parameters. Specifically, as a feasible implementation method, the AI ​​model intelligent review and control platform obtains the condition parameters from the filter conditions and prompt words.

[0037] The business system determines the specific request parameter values ​​based on the condition parameters and sends a second request to the AI ​​model intelligent audit and control platform. The second request carries the request parameter values ​​and the result callback address.

[0038] The AI ​​model intelligent review and control platform matches the request parameter values ​​with a pre-set rule base to determine at least one target AI model and its corresponding prompt words, and adds the AI ​​task to the asynchronous task queue.

[0039] The AI ​​model intelligent review and control platform calls the AI ​​model base and asynchronously executes AI tasks in the task queue to obtain results from at least one target AI model.

[0040] The AI ​​model-based intelligent review and control platform processes the results and returns them to the business system via a result callback address.

[0041] Furthermore, the AI ​​model intelligent review and control platform matches the request parameter values ​​with a pre-set rule base to determine at least one target AI model and its corresponding prompt words, including:

[0042] The request parameter values ​​are matched with multiple predefined rules in the rule pool. The rules define the mapping relationship between business scenarios, judgment conditions, AI models, and prompt words.

[0043] Filter out all matching rules, and obtain the corresponding target AI model and prompt word configuration based on the rules, and package them into a task list.

[0044] Specifically, such as Figure 2 As shown, the AI ​​model intelligent review and control platform receives requests, brings parameter values ​​and extracted information into the rule pool for matching, filters out matching rules, obtains the corresponding prompt word configuration according to the configuration rules, encapsulates it into a list, and passes it into the intelligent queuing and scheduling.

[0045] For example, when parameter information is passed to the rule pool, if a "sensitive word / blacklist" rule that has no conditions and must be executed is matched, it is directly added to the list to be invoked. As another example, during the matching of legal and regulatory rules, if the parsed information indicates that it involves laws and regulations and the contract type is a lease contract, the relevant laws and regulations are matched according to the preset prompt word invocation model, and the task is recorded in the list to be invoked.

[0046] Furthermore, the rules include a first category of rules that must be executed unconditionally and a second category of rules that are triggered based on content characteristics in the request parameter values.

[0047] Furthermore, the implementation of adding AI tasks to the asynchronous task queue includes:

[0048] Check if the current number of queued tasks has reached the threshold;

[0049] If the threshold is not reached, the encapsulated task list will be placed in the specified task queue for scheduling; if the threshold is reached, an error message will be returned.

[0050] Specifically, such as Figure 3 As shown, after receiving a request, the AI ​​model intelligent review and control platform checks whether the request type is configured. If not, it directly returns an error message. It checks whether the number of queued requests has reached the threshold. If it has, it returns an error message. If it has not, it puts the request into the specified queue. The idle client retrieves the target AI model base capability from the queue. The target AI model base capability feedback result is then called back to the business end.

[0051] It's important to note that in the context of a queuing system, a client refers to the entity or agent requesting the service. It represents the individual or component initiating the service request and is responsible for sending the request to the agent. For example, in a scalable queuing system, multiple clients share the server's concurrency capabilities: assuming the server supports 300 concurrent requests, when three clients run simultaneously, each client is allocated 100 concurrent slots. If client A completes 10 tasks, its free slots will automatically receive 10 new tasks from the queue, achieving dynamic load balancing.

[0052] Furthermore, the AI ​​model intelligent review and control platform processes the results by integrating and analyzing multiple results from multiple target AI models when there are multiple target AI models, forming a unified result.

[0053] Furthermore, the intelligent audit and control platform records the entire call log from receiving the second request to returning the result. The log includes the request parameter values, matching rules, the AI ​​model called, the time consumed, the return results of each model, and the final result.

[0054] In summary, by adopting the AI ​​capability invocation method of this invention, the original model, which required code writing and involved code modification, testing, and release processes for each prompt word optimization or rule adjustment, has been transformed into a solution that connects to an intelligent audit and control platform, resulting in an overall efficiency improvement of approximately 70%.

[0055] The AI ​​capability invocation system based on the AI ​​model-based intelligent review and control platform includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the AI ​​capability invocation method based on the AI ​​model-based intelligent review and control platform.

[0056] A computer-readable storage medium storing a computer program, wherein when executed by a processor, the computer program implements the steps of an AI capability invocation method for an AI model-based intelligent audit and control platform.

[0057] The present invention adopts the above technical solution and has the following beneficial effects compared with the prior art:

[0058] 1) The platform centrally manages the access, authentication, monitoring, and alerting of all AI models, eliminating the previous need to configure and maintain multiple model connections separately for each business system, significantly reducing system operation and maintenance complexity and costs. 2) Interaction is achieved through a unified, standard API provided by the platform, completely decoupling from the underlying specific AI models. When adding or replacing AI models or optimizing prompts, no modification to any business system code is required; the entire process can be completed visually through the platform's configuration interface, shortening the development, testing, and deployment cycle from weeks to hours, significantly improving demand response efficiency. 3) Through the dynamic configuration function of rules and prompts provided on the front end, business personnel can quickly respond to policy changes. For newly emerging content risks, new rules can be configured immediately and associated with the corresponding models. 4) The platform provides a queuing function to prevent AI models from experiencing congestion and performance issues during peak periods; it also records complete logs of AI capability calls, providing test cases for subsequent AI optimization; and it provides a large-screen reporting function, fully displaying the effects of AI capability calls.

[0059] This invention employs a microservice architecture to achieve unified access, dynamic configuration, capability orchestration, rate limiting, monitoring logs, and data feedback for model calls. Business systems can invoke various AI capabilities simply through the platform's APIs, without needing to concern themselves with details such as underlying model switching and prompt word adjustments. The platform boasts excellent scalability and high performance, fully meeting the call demands during peak business periods.

[0060] Obviously, the described embodiments are only a part of the embodiments of this application, not all of them. Without conflict, the embodiments and features in the embodiments of this application can be combined with each other. The components of the embodiments of this application described and illustrated herein can generally be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of this application is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

Claims

1. A method for invoking AI capabilities based on an AI model-based intelligent review and control platform, applied to a system including business systems and the AI ​​model-based intelligent review and control platform, characterized in that: The methods include: The business system initiates the first request to the AI ​​model-based intelligent review and control platform; The AI ​​model intelligent review and control platform returns pre-configured condition parameters for the AI ​​task to the business system based on the first request. The condition parameters include filterable conditions and prompt word template parameters. The business system determines the specific request parameter values ​​based on the condition parameters and sends a second request to the AI ​​model intelligent audit and control platform. The second request carries the request parameter values ​​and the result callback address. The AI ​​model intelligent review and control platform matches the request parameter values ​​with a pre-set rule base to determine at least one target AI model and its corresponding prompt words, and adds the AI ​​task to the asynchronous task queue. The AI ​​model intelligent review and control platform calls the AI ​​model base and asynchronously executes AI tasks in the task queue to obtain results from at least one target AI model. The AI ​​model-based intelligent review and control platform processes the results and returns them to the business system via a result callback address.

2. The AI ​​capability invocation method based on the AI ​​model-based intelligent audit and control platform according to claim 1, characterized in that, The AI ​​model intelligent review and control platform matches the request parameter values ​​with a pre-set rule base to determine at least one target AI model and its corresponding prompt words, including: The request parameter values ​​are matched with multiple predefined rules in the rule pool. The rules define the mapping relationship between business scenarios, judgment conditions, AI models, and prompt words. Filter out all matching rules, and obtain the corresponding target AI model and prompt word configuration based on the rules, and package them into a task list.

3. The AI ​​capability invocation method based on the AI ​​model-based intelligent audit and control platform according to claim 1, characterized in that... The rules include the first type of rules that must be executed unconditionally and the second type of rules that are triggered based on the content characteristics in the request parameter values.

4. The AI ​​capability invocation method based on the AI ​​model-based intelligent audit and control platform according to claim 1, characterized in that, The implementation of adding AI tasks to the asynchronous task queue includes: Check if the current number of queued tasks has reached the threshold; If the threshold is not reached, the encapsulated task list will be placed in the specified task queue for scheduling; if the threshold is reached, an error message will be returned.

5. The AI ​​capability invocation method based on the AI ​​model-based intelligent audit and control platform according to claim 1, characterized in that... The AI ​​model intelligent review and control platform processes the results, including: when there are multiple target AI models, it integrates and analyzes multiple results from multiple target AI models to form a unified result.

6. The AI ​​capability invocation method based on the AI ​​model-based intelligent audit and control platform according to claim 1, characterized in that... The intelligent audit and control platform records the entire call log from receiving the second request to returning the result. The log includes the request parameter values, matching rules, the AI ​​model called, the time consumed, the return results of each model, and the final result.

7. An AI capability retrieval system based on an AI model-driven intelligent audit and control platform, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes a computer program, it implements the steps of the AI ​​capability invocation method of the AI ​​model-based intelligent audit and management platform according to any one of claims 1 to 6.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the AI ​​capability invocation method of the AI ​​model-based intelligent audit and control platform according to any one of claims 1 to 6.

Images