KR102785518B1 - Application management module and method using AI-based small language model and retrieval augmented generation technology - Google Patents

Abstract

The application management module and management method using the artificial intelligence-based small language model and the search augmentation generation technology according to the present invention have the advantage of allowing even an administrator who lacks specialized IT knowledge to input a query about the operating status of an application in natural language and receive a response to the query in natural language, thereby enabling more easy and efficient control and management of the application. In addition, since it is configured to use the small language model and the search augmentation generation module, it is possible to generate a more accurate response based on the query and the search result as well as the search result of searching the monitoring data stored in the database in response to the administrator's query, thereby enabling more accurate and reliable responses to be derived about the operating status of the application. In addition, since the present invention is configured to use a small language model having a relatively small number of parameters compared to a conventional large language model, it can be applied to low-spec computing devices without a GPU, etc., thereby enabling the advantage of easier monitoring of the application.

Description

{Application management module and method using AI-based small language model and retrieval augmented generation technology}

The present invention relates to an application management module and a management method using an artificial intelligence-based small language model and a search augmentation generation technology, and more specifically, to an application management module and a management method using an artificial intelligence-based small language model and a search augmentation generation technology, which enable an administrator to control an application and monitor its operating status more easily.

In general, generative AI is a field of artificial intelligence that can generate new content and ideas, such as text, images, and media, in response to prompts in natural language, which is human language. It can learn natural language, programming languages, art, and academia, and reuse learning data to solve new problems or create new content.

This generative AI is based on a large language model (LLM), which is an artificial neural network with billions of parameters that learns a huge amount of data and has the ability to understand and generate natural language.

However, since conventional generative AI is based on large language models (LLMs), it requires a huge amount of computing resources, and monitoring and managing the application has limitations in that it has to rely heavily on the experience of professional engineers or managers.

Korean Patent No. 10-2637029

The purpose of the present invention is to provide an application management module and management method using an artificial intelligence-based small language model and search augmentation generation technology that can control and monitor applications more easily and conveniently.

An application management module using an artificial intelligence-based small language model and a search augmented generation technology according to the present invention includes: an embedding module for converting monitoring data on the operation of an application to be managed into an embedding vector; a vector database in which the embedding vectors for the monitoring data converted by the embedding module are stored; a prompt interface for receiving a query in natural language about the operation status of the application from an administrator terminal possessed by an administrator who monitors the application during its operation; and a search augmented generation (RAG) module for searching for data related to the query in the vector database, inputting both the query and the search data retrieved from the vector database into a small language model (SLM) pre-trained by artificial intelligence, and generating a response according to the query and the search data in natural language from the small language model.

The above prompt interface may be based on Hypertext Transfer Protocol (HTTP).

The above prompt interface can transmit a response derived from the search augmentation generation module to the administrator terminal.

The above monitoring data may include data on the management target function of the application and the operating status of devices including CPU, memory, and connection devices for performing the application.

The above-mentioned embedding module can convert the collected data by periodically polling the file database in which the above-mentioned monitoring data is stored, and the application can convert the collected data by periodically calling the API (Application Programming Interface).

The above response may include at least some of the operating status of the application, the cause of the malfunction, a solution, and the processing result.

The application management method using an artificial intelligence-based small language model and a search augmentation generation technology according to the present invention may include a vector storage step in which an embedding module converts monitoring data on the operation of an application into an embedding vector and stores the same in a vector database; a query reception step in which a prompt interface receives a query in natural language about the operation status of the application from an administrator terminal possessed by an administrator who monitors the application during the operation of the application; a search step in which a search augmentation generation module searches the vector database for data related to the query; and a response derivation step in which the search augmentation generation module inputs the query received in the query reception step and the search data searched in the search step into a small language model (SLM) pre-learned by artificial intelligence, and derives a response in natural language according to the query and the search data from the small language model.

The above monitoring data may include data on the management target function of the application and the operating status of devices including CPU, memory, and connection devices for performing the application.

The above-mentioned embedding module can convert the collected data by periodically polling the file database in which the above-mentioned monitoring data is stored, and the application can convert the collected data by periodically calling the API (Application Programming Interface).

The above prompt interface may be based on Hypertext Transfer Protocol (HTTP).

The method may further include a response output step of outputting the response derived in the above response derivation step to the administrator terminal through the prompt interface.

The above response may include at least some of the operating status of the application, the cause of the malfunction, a solution, and the processing result.

The application management module and management method using the artificial intelligence-based small language model and search augmentation generation technology according to the present invention have the advantage of allowing even managers lacking specialized IT knowledge to input queries about the operating status of applications in natural language and receive responses to the queries in natural language, thereby enabling easier and more efficient control and management of applications.

In addition, since it is configured to utilize a small language model and a search augmentation generation module, it is possible to generate a more accurate response based on the query and the search result as well as the search result of searching the monitoring data stored in the database for the manager's query, so there is an advantage in that a more accurate and reliable response can be derived about the operating status of the application.

In addition, since the present invention is configured to use a small language model with a relatively small number of parameters compared to conventional large language models, it can be applied to low-spec computing devices without a GPU, etc., and thus has the advantage of allowing easier monitoring of applications.

FIG. 1 is a block diagram schematically illustrating the configuration of an application management module using an artificial intelligence-based small language model and search augmentation generation technology according to an embodiment of the present invention.
FIG. 2 is a diagram schematically illustrating the collection and conversion of monitoring data in an application management module according to an embodiment of the present invention.
FIG. 3 is a flowchart illustrating an application management method using an artificial intelligence-based small language model and search augmentation generation technology according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described with reference to the attached drawings.

An application management module according to an embodiment of the present invention is a dynamic library module for monitoring and managing the operating status of an application or the operating status of devices required to perform the application.

FIG. 1 is a block diagram schematically illustrating the configuration of an application management module using an artificial intelligence-based small language model and search augmentation generation technology according to an embodiment of the present invention.

Referring to FIG. 1, the application management module according to an embodiment of the present invention includes an embedding module (10), a vector database (20), a prompt interface (30), a search augmentation generation module (40), and a file database (50).

The above embedding module (10) converts monitoring data on the operation of the application to be managed into an embedding vector.

Here, the above application is explained as an example of a generative AI application, but is not limited to this and can be applied to various applications.

The above monitoring data may include data on the management target function of the application and the operating status of devices required to perform the application. The devices may include a CPU, memory, a connection part, etc. However, the above monitoring data is not limited thereto, and the monitoring data may be applied in various ways if it is data that must be monitored for the smooth operation of the application.

Referring to FIG. 2, the monitoring data may be collected by periodically polling the file database (50), and the application may collect changed data by periodically calling an API (Application Programming Interface).

In the above vector database (20), the embedding vectors for the monitoring data converted in the embedding module (10) are stored.

The above prompt interface (30) receives a query about the operation of the application in natural language from an administrator terminal (70) possessed by an administrator who wishes to monitor the application. Here, natural language means human language.

The above query is an interactive message for the administrator to monitor the operating status of the application. The above query may include questions, commands, etc.

The above prompt interface (30) can be provided based on Hypertext Transfer Protocol (HTTP) that can be used by both administrators and systems.

The above prompt interface (30) can transmit a response derived from the search augmentation generation module (40) to the administrator terminal (70) via wired or wireless communication.

The above administrator terminal (70) can be any terminal capable of communication, such as a desktop, laptop, mobile terminal, or tablet PC.

The above-described Retrieval Augmented Generation Module (RAG Module) (40) is a module that searches for monitoring data related to the query among the data stored in the vector database (20), inputs the searched search data and the query into a small language model (sLM, Small Language Model) (60), and generates and derives a response to the query in natural language from the small language model (60).

That is, the search augmentation generation module performs both the function of searching for information from the vector database (20) according to the query and the function of generating a response using the small language model (60) based on the query and the search data.

The above small language model (60) is a language model having a relatively small number of parameters compared to a general large language model (LLM, Large Language Model), and since it uses less memory and has lower computational requirements compared to the large language model, it has the advantage of being usable even on mobile devices without a GPU or low-spec computing devices. However, the present invention is not limited to a small language model, and can also be applied to a general large language model.

The above file database (50) is a database in which the above monitoring data is created and stored.

The application management method using a small language model and search augmentation generation technology according to an embodiment of the present invention configured as described above is described as follows.

FIG. 3 is a flowchart illustrating an application management method using a small language model and search augmentation generation technology according to an embodiment of the present invention.

Referring to FIG. 3, an application management method according to an embodiment of the present invention includes a vector storage step (S1), a query receiving step (S2), a search step (S3), a response derivation step (S4), and a response output step (S5).

The above vector storage step (S1) is a step of collecting monitoring data on the operation of the application, converting the collected monitoring data into an embedding vector, and storing it in the vector database (20).

In the above vector storage step (S1), the embedding module (10) can convert monitoring data collected at a preset cycle or in real time into an embedding vector and then store it.

The above query receiving step (S2) is a step in which the prompt interface (30) receives a query in natural language from the administrator terminal (70) to monitor the operation of the application.

The above query can be made in the form of an interactive message.

For example, the above query can be entered in various ways, such as “Is the application working well now?”, “Tell me the application status every 20 minutes”, or “If a concurrent connection problem occurs, increase the connection pool”, as long as it is a message that can monitor the operating status of the application.

The above prompt interface (30) can also support repetitive or delayed queries through a timer function. For example, the repetitive query can be a message such as “notify me every 20 minutes,” and the delayed query can be a message such as “notify me after 20 minutes.”

The above search step (S3) is a step in which, when the above query is received, the search augmentation generation module (40) searches for monitoring data related to the query among the data stored in the vector database (20).

In the above search step (S3), monitoring data for queries similar to the above query, or monitoring data corresponding to the above query and having a high degree of relevance, etc. can be searched in the vector database (20).

For example, if the above query is “Tell me the current operating status of the application,” monitoring data on the operating status for the past week can be retrieved from the vector database (20).

The above response derivation step (S4) is a step in which the search augmentation generation module (40) inputs the query entered into the prompt interface (20) and the search data searched in the search step (S3) into the small language model (60), and derives a response to the query in natural language from the small language model (60).

In the above response derivation step (S4), an information search function that searches for information stored in the vector database (20) and a response generation function that generates a response using the small language model (60) according to the searched information and the query are performed together, thereby generating a more accurate and useful response. That is, in the response derivation step (S4), a more accurate response to the query can be derived from the small language model (60) by using not only the query but also all of the search data searched according to the query.

The above response may be in the form of an interactive message.

The above response may include at least some of the operating status of the application, the cause of the malfunction, a solution, and the processing result.

For example, the response to a query such as “Is the application working well now?” could be a message including the operating status and the cause of the malfunction, such as “The service is currently not available. The cause seems to be a change in database information.” Also, the response to a query such as “If a concurrent connection problem occurs, increase the connection pool.” could be a message including the operating status and a solution, such as “The current concurrent connections exceed the allowable limit. I will increase the connection pool.” Also, the response to a query such as “Tell me the application status every 20 minutes.” could be a message including the operating status, such as “The application status will be fine in 20 minutes.”

After a response to the above query is derived in the above response derivation step (S4), an additional query including additional instructions or commands for the response can be additionally input from the administrator.

In the above response output step (S5), the response derived in the response derivation step (S4) is output to the administrator terminal (70) through the prompt interface (30).

Although the present invention has been described with reference to the embodiments shown in the drawings, these are merely exemplary, and those skilled in the art will understand that various modifications and equivalent other embodiments are possible therefrom. Accordingly, the true technical protection scope of the present invention should be determined by the technical idea of the appended claims.

10: Embedding Module 20: Vector Database
30: Prompt Interface 40: Search Augmentation Generation Module
50: File database 60: Small language model
70: Admin terminal

Claims (12)

A dynamic library module for monitoring and managing the operating status of an application to be managed or the operating status of devices required to perform said application.
An embedding module included in a runtime search augmentation generation (RAG) library built within the application, which converts monitoring data on the operation of the application into an embedding vector;
A vector database in which the embedding vectors for the monitoring data converted by the embedding module are stored, and which is included in the runtime search augmentation generation library;
A prompt interface based on Hypertext Transfer Protocol (HTTP) for receiving a query in natural language about the operating status of the application from an administrator terminal possessed by an administrator who monitors the application while the application is in operation;
A Retrieval Augmented Generation (RAG) module is included in the runtime search augmented generation library, and retrieves data related to the query from the vector database, inputs both the query and the search data retrieved from the vector database into a small language model (SLM) pre-trained by artificial intelligence, and generates and derives a response according to the query and the search data in natural language from the small language model.
The above monitoring data is,
Contains data on the management target function of the above application and the operating status of the device including the CPU, memory, and connection device for performing the above application,
The above response is,
Contains at least some of the operating status, cause of malfunction, solution, and processing result of the above application,
The above embedding module,
The above monitoring data is collected by periodically polling the file database where it is stored and converting the collected data.
The above application periodically calls the API (Application Programming Interface) to convert the collected data.
Application management module using artificial intelligence-based small language models and search augmentation generation technology. delete In claim 1,
The above prompt interface is,
Transmitting the response derived from the above search augmentation generation module to the above administrator terminal,
Application management module using artificial intelligence-based small language models and search augmentation generation technology. delete delete delete In an application management method using a dynamic library module for monitoring and managing the operating status of an application to be managed or the operating status of devices required to perform the application,
A vector storage step in which the embedding module included in the runtime search augmentation generation (RAG) library built within the above application converts monitoring data on the operation of the application into an embedding vector and stores it in a vector database included in the runtime search augmentation generation library;
A query receiving step in which, during the operation of the above application, a prompt interface based on Hypertext Transfer Protocol (HTTP) receives a query in natural language about the operation status of the above application from an administrator terminal possessed by an administrator monitoring the above application;
A search step in which a search augmentation generation module included in the above runtime search augmentation generation library searches for data related to the query in the above vector database;
The above search augmentation generation module includes a response derivation step in which the query received in the above query receiving step and the search data searched in the above search step are input into a small language model (SLM, Small Language Model) pre-learned by artificial intelligence, and a response according to the query and the search data is derived in natural language from the small language model.
The above monitoring data is,
Contains data on the management target function of the above application and the operating status of the device including the CPU, memory, and connection device for performing the above application,
The above response is,
Contains at least some of the operating status, cause of malfunction, solution, and processing result of the above application,
The above embedding module,
The above monitoring data is collected by periodically polling the file database where it is stored and converting the collected data.
The above application periodically calls the API (Application Programming Interface) to convert the collected data.
Application management method using artificial intelligence-based small language models and search augmentation generation technology. delete delete delete In claim 7,
Further comprising a response output step for outputting the response derived in the above response derivation step to the administrator terminal through the prompt interface.
Application management method using artificial intelligence-based small language models and search augmentation generation technology. delete