Application generation method, program, and apparatus

Abstract

To quickly and easily generate applications with guaranteed quality. [Solution] A method for a computer to generate an application for extracting and visualizing data from an unstructured data file, the method comprising the steps of: the computer referencing a common infrastructure that holds either or both a common code layer containing definitions of general functions and a design reference layer containing implementation patterns suited to specific business purposes; and sending a prompt to an AI agent to generate the application, the prompt comprising: read logic defining rules for extracting data from the unstructured data file; and instructions that refer to at least one of the common code layer and the design reference layer of the common infrastructure in generating the application.

Description

Technical Field

[0006] , , ,

[0005] , , , , ,

[0001] The present invention relates to an application generation method, an application generation program, and an application generation device.

Background Art

[0002] Patent Document 1 describes an invention that generates an application using an external AI system (such as a generation AI system).

[0003] Patent Document 2 describes a configuration that uses pre-defined templates and common modules to generate source code and test code based on each schema.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0005] An object of the present invention is to quickly and easily generate an application with guaranteed certain quality.

Means for Solving the Problems

[0007] In the above method, the instruction to reference the common infrastructure in the prompt may be a reference to the common code layer only, a reference to the design reference layer only, or a reference to both the common code layer and the design reference layer, depending on the purpose of generating the application.

[0008] In the above method, the generated application may, at runtime, display on the terminal's display unit a preview screen that includes an upload screen for the unstructured data file, a data structure defined based on the reading results of the unstructured data file, and a display of the converted table format.

[0009] In the above method, the generated application may, at runtime, receive a modification from the terminal via the preview screen to either the reading logic or the reading position that identifies the matrix range for data extraction on the unstructured data file. If such modification is received, the application may rewrite the parameters of the reading settings without rebuilding the entire application.

[0010] According to another aspect of the present invention, the present invention provides a program that causes a computer to perform any of the above methods.

[0011] According to yet another aspect of the present invention, there is a device for generating an application that extracts and visualizes data from an unstructured data file. The device comprises a control unit, a storage unit, and a communication unit. The storage unit stores a common infrastructure that holds either or both a common code layer containing definitions of general functions and a design reference layer containing implementation patterns suitable for specific business purposes. The control unit refers to the common infrastructure and sends a prompt to an AI agent via the communication unit for generating the application. The prompt includes read logic that defines rules for extracting data from the unstructured data file and an instruction that, in generating the application, refers to at least one of the common code layer and the design reference layer of the common infrastructure. [Effects of the Invention]

[0012] According to the present invention, applications with a certain level of quality can be generated quickly and easily.

[0013] It should be noted that the above summary of the invention does not enumerate all of its features. Furthermore, subcombinations of these features may also constitute an invention. [Brief explanation of the drawing]

[0014] [Figure 1] This is a system configuration diagram relating to one embodiment of the present invention. [Figure 2] This is a functional configuration diagram of Server 1 according to one embodiment of the present invention. [Figure 3] This is a functional configuration diagram of terminal 2 according to one embodiment of the present invention. [Figure 4] This is a conceptual diagram of the structure of the common platform 130 and prompt related to one embodiment of the present invention. [Figure 5] This figure shows the processing flow during the generation of a generation application 150 according to one embodiment of the present invention. [Figure 6] This figure shows the execution process flow of a generation application 150 according to one embodiment of the present invention. [Modes for carrying out the invention]

[0015] The present invention will be described in detail below with reference to the drawings illustrating its embodiments. Note that the following embodiments are not intended to limit the invention as described in the claims. Furthermore, not all combinations of features described in the embodiments are necessarily essential as means of solving the invention.

[0016] Figure 1 is a system configuration diagram relating to one embodiment of the present invention. The system according to this embodiment is an application generation system consisting of a server 1, a terminal 2, an AI agent server 3, and a network N, and Figure 1 is a conceptual diagram showing its overall configuration. This system is for dynamically constructing and executing an application (generation application 150, hereinafter also simply referred to as "application 150") that extracts and visualizes data from unstructured data files. In this specification, "system" refers to a configuration consisting of one or more computers, servers, terminals, devices, control units, storage units, communication units, functional units, or any combination thereof, which performs predetermined information processing. The system may be configured as a single physical device, or it may be configured distributed among multiple physical devices. Furthermore, the system may consist of multiple devices connected via a network as shown in Figure 1, or it may be implemented in a virtual environment or a cloud environment.

[0017] Here, in FIG. 1, the system according to this embodiment includes a server 1 responsible for central control, one or more terminals 2 operated by developers or users, and an AI agent server 3 that functions as an external resource. These are connected to be mutually communicable via a network N such as the Internet or an intranet. In FIG. 1, the terminal 2 is shown as including a developer terminal 2a and a user terminal 2b, but the terminal 2 may be composed of only any one of these, or may include other terminals. Hereinafter, without distinguishing between the terminal 2a, the terminal 2b, and other terminals, the terminal used in this embodiment is also simply described as the "terminal 2". Also, in this specification, the developer terminal 2a may be simply described as the "terminal 2a", and the user terminal 2b may be simply described as the "terminal 2b".

[0018] In this embodiment, the server 1 has the functions of a device that generates an application 150 for extracting and visualizing data from unstructured data files, and an execution device that operates the application 150. The server 1 is composed of a computer. Here, in this specification, the "computer" refers to an information processing device that includes a processor and a storage device, and the processor executes a program stored in the storage device to perform predetermined information processing. The computer may be configured as a single physical device, or may be realized as a distributed configuration in which a plurality of physical devices or virtual devices cooperate via a network. Note that the computer may include, but is not limited to, a server, a personal computer, a virtual machine, a processing device in a cloud computing environment, a mobile terminal, an embedded device, or any combination thereof. Also, in this specification, the "server" refers to a computer that receives requests from other computers via a network and executes processing corresponding to the requests. Note that in this specification, the "server" and the "terminal" indicate functional roles, and are not limited to being physically separate devices. They may be realized by the same computer, or may be distributedly realized by a plurality of computers.

[0019] As used herein, the term "unstructured data file" refers to a file containing data whose composition, arrangement, name, meaning, or format is not fixedly defined and may have a different structure for each file. Unstructured data files may include, for example, spreadsheet files (such as Excel (registered trademark) files), document files, text files, structured data files (such as JSON, XML, and CSV), semi-structured data files, and unstructured data files, but are not limited thereto. Further, as used herein, the term "application for extracting and visualizing data from an unstructured data file" refers to a program, information processing module, or information processing apparatus having a function of acquiring data included in an unstructured data file, extracting predetermined data by analysis processing, identification processing, or extraction processing, and visualizing the extracted data by display, charting, structured display, or other visual representations. In the present embodiment, the application is shown as a generation application 150 in FIG. 2. The application can analyze the structure of an unstructured data file, perform item identification, classification, association, conversion, or arrangement, and display the result via a user interface. Further, the extraction processing may be executed by rule-based processing, statistical processing, machine learning processing, or any combination thereof, but is not limited thereto. The function of the server 1 as an application generation apparatus for generating an application 150 for extracting and visualizing data from an unstructured data file will be described in detail using FIG. 2 below.

[0020] Note that the server 1 in the present embodiment has at least two different operating states: a generation stage that leads the construction of an application and an execution stage that operates the constructed application, depending on its operation phase.

[0021] First, during the application generation phase, the control unit 110 (application generation unit 112) of server 1 functions as a generation device. Specifically, it integrates individual requirements (read definition data 140) based on unstructured data files received from terminal 2 with the common infrastructure 130 held in the storage unit 120, and sends a prompt to AI agent server 3. The main role of server 1 at this stage is to receive information such as source code returned from the AI ​​agent and to build and deploy a generated application 150 optimized for a specific business purpose in an executable state.

[0022] On the other hand, in the application execution phase after the generated application 150 has been built, server 1 functions as an application server (execution platform) for running the generated application 150. At this stage, the control unit 110 (application execution unit 114) of server 1 loads the previously generated program into memory and executes it. Specifically, processes such as receiving uploads of actual data from terminal 2, structuring and converting data based on read definition data 140, storing it in a database, and generating preview screens such as dashboards are all performed under the control of this generated application 150. Furthermore, if the server 1 receives parameter modifications from a developer or user via terminal 2 during execution, it dynamically rewrites the operation settings of the running application 150, immediately updating the processing content without going through the regeneration process.

[0023] In this way, Server 1 seamlessly switches between the stage of generating Application 150 in cooperation with the AI ​​agent and the stage of performing specific data processing using the generated Application 150, thereby providing a consistent service from development to operation and even fine-tuning in the field. Hereafter, when Server 1 operates based on Application 150, it may simply be described as the operation of Application 150.

[0024] Terminal 2 is a computer involved in the generation and execution of application 150 by server 1. Here, the description of a computer is omitted as it overlaps with the description of server 1 above, but in this specification, "terminal" refers to a computer that communicates with a server via a network, sends requests to the server, or receives processing results from the server.

[0025] In this embodiment, terminal 2 may include a developer terminal 2a as a terminal used by a developer, and a user terminal 2b as a terminal used by a user. The developer terminal 2a is primarily involved in the generation of application 150 by server 1, but is not limited to this, and may also be involved in the execution of application 150. For example, a developer may generate application 150 by server 1 via the developer terminal 2a, while also using the developer terminal 2a as a user for testing and production operation of the generated application 150. Similarly, the user terminal 2b is primarily involved in the execution of application 150 by server 1, but is not limited to this, and may also be involved in the generation of application 150. For example, a user may access server 1 via user terminal 2b and execute and use the generated application 150, while also generating application 150 by providing predetermined input to user terminal 2b.

[0026] AI Agent Server 3 is a computer that enables Server 1 to use AI agents in generating Application 150. Hereinafter, "AI agent" means an information processing entity that has the function of performing inference, judgment, or decision-making based on input information and autonomously executing predetermined processing according to the result. An AI agent may include, for example, at least one of a machine learning model, rule-based inference unit, neural network, language model, large-scale language model (LLM), reinforcement learning model, orchestrator, multi-agent system, inference engine, polyphonic AI, or a combination thereof. An AI agent may also have the function of sending and receiving information with external devices, other programs, databases, sensors, or user interfaces, determining processing content based on such information, and executing said processing.Hereinafter, an AI agent made available by AI Agent Server 3 will also be simply referred to as "AI agent".

[0027] Network N is a communication network that connects multiple information processing devices in a communicative manner, enabling the transmission and reception of data between these devices. Network N may include, but is not limited to, the Internet, a local area network (LAN), a wide area network (WAN), a cloud network, a virtual network, a virtual private network (VPN), a wireless network, a mobile communication network, and a closed network. Furthermore, Network N may be configured using either wired or wireless communication, may be configured as a single network, or may be configured as multiple networks interconnected to one another.

[0028] Figure 2 is a functional configuration diagram of server 1 according to one embodiment of the present invention. In this embodiment, server 1 comprises a control unit 110, a storage unit 120, and a communication unit 160.

[0029] The control unit 110 includes an application generation unit 112 and an application execution unit 114. Hereinafter, "control unit" means an information processing function that includes a processor, microprocessor, CPU, or other arithmetic processing unit and a program executed by said arithmetic processing unit, and has the function of controlling the operation of the device by executing predetermined arithmetic processing, decision processing, control processing, or information processing. The control unit may be implemented by a single processor, implemented in a distributed manner by multiple processors, or implemented in a virtual processing environment. Furthermore, the control unit may be implemented by dedicated hardware, a program, or any combination thereof.

[0030] The application generation unit 112 refers to an information processing function that generates an application 150 for processing and / or visualizing data based on data extracted from an unstructured data file, processing content corresponding to said data, and / or prompts defining said processing content. The application generation unit 112 may generate programs, configuration information, processing definitions, screen definitions, display definitions, or any combination thereof that constitute the application 150 based on the prompts. Here, "prompt" refers to instruction information for defining the processing content, processing conditions, display content, display format, data extraction conditions, data conversion conditions, or processing specifications including these, for the application. The application generation unit 112 may be implemented by a processor executing a program, by dedicated hardware, or by a combination thereof.

[0031] The application execution unit 114 is an information processing function that executes the generated application 150 and performs processing, display, visualization, input reception, and information processing based on data extracted from an unstructured data file. The application execution unit 114 may read and execute the application 150 generated by the application generation unit 112, or it may execute an application obtained from an external device, or it may execute a combination of these applications.

[0032] The storage unit 120 stores the common base 130, read definition data 140, and generation application 150. Hereinafter, “storage unit” means a storage device or storage function capable of storing programs and / or data. The storage unit may include, but is not limited to, semiconductor memory, RAM, ROM, non-volatile memory, flash memory, hard disk, solid-state drive, optical disc, or any combination thereof. The storage unit may be configured as a single storage device or distributed among multiple storage devices. The storage unit may also be located inside the device or configured as an external storage device accessible via a network.

[0033] The common platform 130 is a data structure that holds a set of information that the AI ​​agent should refer to when generating structured data suitable for specific business purposes from unstructured data files, and when generating an application 150 for visualizing said structured data. In this embodiment, the common platform 130 may include a common code layer 132 that defines general-purpose functions or processes in the program, and a design reference layer 134 that defines implementation patterns in specific success cases or business domains. By having two layers with different reusability characteristics, the common platform 130 can achieve both quality and flexibility in code generation by the AI ​​agent.

[0034] Here, the common code layer 132 may define a set of standard code that can be reused across multiple applications, such as table display, data validation, or file upload. The common code layer 132 is primarily intended for code reuse. For example, general-purpose program code that would result in an enormous number of lines of code if generated from scratch, making AI-based quality control difficult, such as complex table display or file upload functions, may be defined as the common code layer 132.

[0035] On the other hand, the design reference layer 134 may define preferred models for specific designs, such as linking rules between specific data items or user interface layouts. The design reference layer 134 is primarily intended for design reuse. For example, in specific use cases such as customer management or inventory management, the design reference layer 134 may retain definitions of data structures that have been successfully built in the past, as well as specific implementation examples (sample code sets) that include optimized UI / UX design patterns such as the arrangement of tabs on upload and preview screens and operation fields. The design reference layer 134 may also retain implementation patterns that include normalized table structures. These may encompass advanced design philosophies that can withstand actual operation while excluding confidential information.

[0036] As an example of a reference instruction for the common platform 130, the server 1 may, in prompting the AI ​​agent, present a specific implementation example included in the design reference layer 134 as a design type, while instructing it to refer to the code in the common code layer 132 for specific processing. In other words, the common platform 130 may hierarchically provide the AI ​​agent with the reuse of functions from the common code layer 132 and the reuse of designs from the design reference layer 134. This makes it possible to have the AI ​​agent output a flexible implementation tailored to the purpose each time, while inheriting a proven design philosophy, unlike conventional methods that simply combine existing libraries (lack of flexibility) or methods that simply replicate past examples (difficulty in extending functionality). Furthermore, even when a file with an unstructured data structure is input, it is possible to support the construction of a consistent generated application 150 that can withstand actual operation with minimal modification instructions.

[0037] Furthermore, the common platform 130 may function not merely as a code library, but as output constraints and quality guidelines for the AI ​​agent. Specifically, the common code layer 132 may define functions such as data transformation processing and validation that can be used in common across multiple applications. In contrast, the design reference layer 134 may define data structures optimized for specific business operations (e.g., sales management or inventory management), and implementation models including user interface placement patterns. Based on reference instructions to either or both of these layers included in the prompt, the AI ​​agent can generate an application 150 without structural inconsistencies by building upon proven patterns as a framework and adding individual requirements (such as reading logic 142), rather than generating code from scratch.

[0038] Furthermore, the common platform 130 functions as a technical constraint (guardrail) when the AI ​​agent generates code. The prompt may include an identifier for inheriting a specific component or design pattern within the common platform 130, and the AI ​​agent may generate source code in a form that calls program code defined in the common code layer 132 based on that instruction. This, unlike technologies that simply retain reusable parts, appropriately limits the degree of freedom of the AI ​​agent's output and dynamically ensures structural consistency that can withstand actual operation.

[0039] The read definition data 140 is a set of information that defines conversion rules for adapting the unstructured data set contained in the unstructured data file to the data structure defined in the common infrastructure 130. Specifically, the read definition data 140 may include a read logic 142 that defines processing procedures and data definitions for converting the extracted data into structured data, and a read position 144 that specifies the range of data to be extracted within the unstructured data file.

[0040] Here, the reading logic 142 defines the execution program for data type conversion, string manipulation, data definition, or calculation formulas of the extracted data. The reading logic 142 may be set by a developer or user via terminal 2, or it may be automatically generated by an AI agent while referring to the common base 130. The reading position 144 includes coordinate information indicating the starting row, starting column, or specific data range of a cell in a spreadsheet file, and is configured to be dynamically changeable via the user interface during the execution of application 150.

[0041] In this way, the read definition data 140 maintains the read logic 142, which defines the data conversion procedure and data definition, and the read position 144, which defines the location of the data, while keeping them separate and interconnected. This allows for the AI ​​agent to regenerate the read logic 142 in response to more complex changes in the file structure, while enabling rapid responses to minor changes in the file structure by simply modifying the read position 144. Thus, it can function as a flexible data conversion platform.

[0042] In this specification, "reading logic" refers to logical rules for extracting specific information from an unstructured data file (for example, extraction conditions such as searching for a specific string and obtaining data at its relative position, and definitions of how the extracted data is handled in each database). On the other hand, "reading position" refers to coordinate information such as the matrix range specifically referenced in the reading logic, and the starting cell position of the data. In this embodiment, these logical rules and specific coordinate information are separated in the instructions (prompts) given to the AI ​​agent. This makes it possible to generate a highly flexible application 150 that can adjust for minor layout differences between files without having to rebuild the entire program, while maintaining the user interface structure of the design reference layer 134 that has been created for each business purpose.

[0043] The generation application 150 is an executable program dynamically generated by the AI ​​agent based on the input unstructured data file and read definition data 140, while referencing at least one of the common code layer 132 and the design reference layer 134, which are included in the common platform 130. Specifically, the generation application 150 may include an upload function for importing the unstructured data file, a data processing function for extracting structured data from the unstructured data file and storing it in a database, and a dashboard display function for visualizing the data in the database in a predetermined format.

[0044] Here, the generated application 150 has a common execution platform in both the prototype environment during the development phase and the production environment (product) after development is complete, and the operation and display content confirmed during the prototype phase can be immediately put into production operation without going through a redevelopment process. Furthermore, the generated application 150 may have a self-setting function that immediately reflects the changes made by the user to the reading position 144 during execution. In this way, the generated application 150, through the combination of code generation technology by an AI agent and a hierarchical reuse layer, enables even non-engineers to build and operate it continuously in a short period of time with minimal work.

[0045] The communication unit 160 is a communication interface that transmits and receives various types of data to and from terminal 2 or AI agent server 3 via a network N such as the Internet. Here, in this specification, "communication unit" refers to a communication function that transmits and / or receives data to and from other information processing devices, servers, terminals, storage devices, or networks. The communication unit may communicate by wired communication, wireless communication, or any combination thereof. The communication unit may consist of a single communication interface or multiple communication interfaces.

[0046] In this embodiment, the communication unit 160 may receive instructions from the terminal 2 regarding unstructured data files, reading logic 142, or reading position 144, and output such information to the control unit 110. The communication unit 160 may also send prompts generated by the control unit 110 to the AI ​​agent server 3 and receive information such as the source code of the generated program returned by the AI ​​agent server 3. Furthermore, the communication unit 160 may have the function of sending data constituting the operation screen of the generated application 150 running in the application execution unit 114 to the terminal 2, or receiving parameters such as application modification instructions input from the terminal 2. In this way, the communication unit 160 functions as an information aggregation point within this system, enabling real-time information synchronization with external devices and the provision of processing results.

[0047] Figure 3 is a functional configuration diagram of terminal 2 according to one embodiment of the present invention. Terminal 2 comprises a control unit 210, a storage unit 220, a communication unit 230, and a display unit 240.

[0048] The control unit 210 may be composed of a computing device including one or more processors such as a CPU (Central Processing Unit), MPU (Micro Processing Unit), GPU (Graphics Processing Unit), or Application Specific Integrated Circuit (ASIC). The control unit 210 comprehensively controls the overall operation of terminal 2 by executing software such as an operating system or various application programs stored in the storage unit 220. Specifically, the control unit 210 performs information transmission and reception processing with server 1 via network N through a web browser, a dedicated client application, or other communication software interface. In addition, the control unit 210 performs processing to display a file upload screen or preview screen on the display unit 240 based on the execution code or rendering data of the generated application 150 received from server 1, and is responsible for analyzing modification instructions or selection operations from the user received via an input unit (not shown) and sending requests corresponding to those operations to server 1. Note that some of the processing by the control unit 210 may be performed by other computing resources on the network.

[0049] The storage unit 220 is composed of one or more storage devices, including at least one of volatile memory such as RAM (Random Access Memory) and non-volatile storage devices such as ROM (Read Only Memory), HDD (Hard Disk Drive), SSD (Solid State Drive), or flash memory. The storage unit 220 holds various programs executed by the control unit 210 (such as operating systems and web browsers), and also temporarily or continuously holds resource data (such as HTML, CSS, JavaScript®, and image data) that constitute the operation screen of the generated application 150 received from the server 1, as well as unstructured data files specified or entered by the user. Furthermore, the storage unit 220 also functions as a buffer area for temporarily storing cache data generated during communication with the server 1, intermediate data related to correction instructions made by the user on the preview screen, and log data indicating the operating status of terminal 2. Furthermore, the memory unit 220 is not limited to those built into the casing of the terminal 2, but may include any memory resource that can be logically accessed from the terminal 2, such as external storage connected via an external interface, or network storage or cloud storage accessible via a network.

[0050] The communication unit 230 is a communication interface for bidirectional data communication with the server 1 via a network N including the Internet, intranet, dedicated line, or mobile communication network (LTE, 5G, etc.). The communication unit 230 may be composed of, for example, a LAN card, a wireless communication module, or a short-range wireless communication (Bluetooth®, etc.) interface circuit and antenna. Under the control of the control unit 210, the communication unit 230 performs the upload process of an unstructured data file specified by the user. The communication unit 230 also sends control signals to the server 1, including correction instructions for fine-tuning the reading position 144 (matrix coordinates, etc.) or changing the reading logic 142 (extraction conditions, data types, etc.) entered in the preview screen of the generation application 150. Furthermore, the communication unit 230 sequentially receives structured data, visualized graph or chart drawing data, or executable code (scripts, binaries, etc.) that defines the operation of the application as the execution result of the generation application 150 constructed or updated on the server 1. Furthermore, the communication protocol used by the communication unit 230 is not limited to HTTP / HTTPS; real-time communication protocols such as WebSocket or proprietary communication protocols may be used, and the transmitted data may be encrypted or compressed.

[0051] The display unit 240 is comprised of a display panel including a liquid crystal display (LCD), an organic light-emitting diode (OLED), or other display device, and is an interface that presents visual information to the user based on control signals from the control unit 210. The display unit 240 may be a display built into the terminal 2, or it may be an externally connected display device such as a monitor or projector. Specifically, the display unit 240 displays the operation screen of the generation application 150 provided by the server 1. This operation screen includes an upload screen for specifying and sending unstructured data files. The display unit 240 also displays a preview screen based on the processing results from the server 1, including the structure of the data extracted from the unstructured data file, the converted table-formatted data, and graphs that visualize the data. Furthermore, the preview screen displayed on the display unit 240 not merely presents information, but also functions as a graphical user interface (GUI) to prompt correction of the reading logic 142 or the reading position 144. The display unit 240 updates its display content in real time in conjunction with user input operations (for example, dragging a displayed row / column range or selecting from a pull-down menu), providing an environment in which the user can intuitively confirm and adjust the consistency between the unstructured data file and the extraction settings.

[0052] Although not shown in this figure, terminal 2 may be equipped with an input unit that accepts instructions or information input from a developer or user, or it may have an input function by being connected to an external input device via a wired or wireless connection. The input unit or input device (hereinafter also simply referred to as "input unit") may be a physical input device such as a keyboard, mouse, touch panel, touchpad, trackball, or pen tablet, or it may be an audio input device such as a microphone, an eye-tracking input device using a camera, or a gesture input device. Terminal 2 may be configured by arbitrarily combining multiple of these input means. Furthermore, the input unit is not limited to being integrated with the housing of terminal 2, but may be an external device connected via a network or input / output interface (USB, Bluetooth®, etc.), or it may be implemented as a logical interface that receives control signals transmitted from another computer. This makes it possible for developers or users to intuitively perform operations such as correcting the reading position on the preview screen, or selecting a file format, via any input means.

[0053] In this specification, when a developer or user performs a predetermined operation via terminal 2 (developer terminal 2a or user terminal 2b), it means that a signal or data generated as a result of the operation is transmitted to server 1 via network N, and server 1 receives the signal or data and performs information processing corresponding to the operation. As an example, the process when a developer uses terminal 2a to obtain a screenshot image of a preview screen, etc., and uses the screenshot image to generate a prompt in server 1 will be described. First, the control unit 210 of terminal 2a captures the display content of the display unit 240 as image data in response to an operation by the developer on the input unit (not shown) and temporarily stores it in the storage unit 220. Next, when the developer performs an operation to embed the image in a prompt on the management screen of server 1, etc. (upload or drag-and-drop operation, etc.), the communication unit 230 transmits the image data to server 1. The application generation unit 112 of server 1 analyzes the received image data and dynamically incorporates the layout, table structure, or specific text information contained in the image as components of a prompt to be transmitted to the AI ​​agent. Thus, the information processing of this embodiment also includes a series of processes in which, starting with a physical operation via terminal 2 by a developer or user, server 1 takes in unstructured data in the form of an image, converts it into instruction data called a prompt, and reflects it in application generation. In this specification, developers and users are not distinguished and are sometimes referred to collectively simply as "users."

[0054] Figure 4 is a conceptual diagram of the structure of the common platform 130 and prompts according to one embodiment of the present invention. Figure 4 is a conceptual diagram showing an example of the stacked structure of the common platform 130 and the configuration of prompts fed to the AI ​​agent, which are the core of the generation of the application 150 in the system of this embodiment.

[0055] The common infrastructure 130 may be conceptually represented as a stacked layer structure in which a common code layer 132 and a design reference layer 134 are stacked from the bottom to the top, as shown in Figure 4. In this figure, the lower layers, which form the foundation, are shown to be more versatile and to have a relatively smaller amount of code (number of data lines, etc.) to hold, but this is not limited to this, and the code capacity of the common code layer 132 may be greater than the code capacity of the design reference layer 134. Here, the bottommost common code layer 132 defines a set of reusable programs that form the basis of the generated applications 150, and the uppermost design reference layer 134 may define implementation samples of designs that show how to combine the common code layers 132 for a specific business purpose. Furthermore, conceptually, one or more generated applications 150 that are generated by referencing these may be located in the upper layers, and when generating other generated applications 150, information from each of these layers, including existing generated applications 150, may be used.

[0056] The prompt may include a reference to the common infrastructure 130, in addition to individual instruction information regarding both or either the reading logic 142 and the reading position 144. By configuring the prompt in this way, combining the individual requirements of the reading logic and position with predefined common code and design samples, the AI ​​agent can construct the generated application 150 while dynamically generating high-quality program code that does not contradict existing design assets. Server 1 can generate the application 150 by utilizing the AI ​​agent's reasoning capabilities even if the prompt does not include individual instruction information regarding both or either the reading logic 142 and the reading position 144. Specifically, even if the prompt does not include explicit instructions specifying the reading position 144 (matrix coordinates, etc.), the AI ​​agent will identify or infer the location of the data within the unstructured data file based on the reading logic 142 included in the prompt (description of the item name to be extracted, data type, or logical structure of the data, etc.). For example, if the reading logic 142 instructs the extraction of the "item master," the AI ​​agent may search the unstructured data file for a column containing that name or a similar string and automatically determine the appropriate reading position 144. Furthermore, when an unstructured data file is uploaded at runtime, the generated application 150 may analyze the contents of the file, dynamically identify or infer data ranges not specified in prior prompts, and perform data extraction.

[0057] Here, the prompt sent by Server 1 may be dynamically configured. Specifically, in addition to metadata obtained from the analysis results of the unstructured data file, it may include an identifier that specifies which component in the common infrastructure 130 should be inherited. Through this mechanism, the output results of the AI ​​agent will take the form of inheriting specific class names and function names defined in the common infrastructure 130. Therefore, by checking the generated source code or analyzing the operation characteristics at runtime, it is possible to determine whether or not the common infrastructure of the present invention has been referenced.

[0058] Furthermore, this system may have a configuration that dynamically switches the reference destination of the common infrastructure 130 depending on the generation purpose, or the reference destination of the common infrastructure 130 may be selected by the developer or user via the operation of terminal 2. For example, if the goal is to completely eliminate coding (e.g., SaaS) for a specific product, an application 150 may be generated that references only the common code layer 132 and operates only by changing settings. On the other hand, when applying to a new domain, past examples of the design reference layer 134 may be used as reference models. In addition, in design patterns on the design reference layer 134 that have been created, parts that have high commonality and a large number of lines of code may be sequentially promoted to the common code layer 132 and incorporated as reusable components. Also, the reference instruction to the design reference layer 134 may include an instruction to map and store data extracted from an unstructured data file in a defined table structure. With such a self-evolving infrastructure configuration, the development efficiency using AI agents can be continuously improved by gradually increasing or decreasing the reused parts.

[0059] Herein, an example of a prompt in this embodiment is shown. For example, a prompt that server 1 sends to AI agent server 3 may include multifaceted instruction data to define the specifications of the generated application 150. Specifically, the prompt may include the following information (1) to (3).

[0060] (1) Instructions regarding the application's purpose: This information defines the problems the application to be generated should solve and the expected use cases. For example, it may include natural language instructions that identify the application's functional role and purpose.

[0061] (2) Instructions regarding the content of the data to be handled: This is information that defines the data items (columns) and their attributes that should be maintained and managed within the application. For example, in addition to the name of the specific data structure, it may include multiple item definitions specific to the business domain, such as the IDs and names of each item included in the sheet.

[0062] (3) Instructions for adding sheets: These are execution instructions that define the screen configuration of the application and the expansion of data storage units. For example, they may include instructions that explicitly prompt the creation of new data management units (tabs or tables). This allows the AI ​​agent to dynamically construct the optimal UI components and database structure (such as a normalized table structure) for holding the specified data items, in accordance with the format of design reference layer 134.

[0063] Furthermore, an example of how to reference the common infrastructure 130 and construct the application 150 in this system will be described in detail. In this example, a common code layer 132 is built within the common infrastructure 130, which defines the standard functions and core logic that form the basis of the application. In addition, a design reference layer 134 optimized for a specific business domain is provided, and it is assumed that the common code layer 132 has already been imported and defined within the design reference layer 134. Here, when generating an application 150 that is suited to a specific business purpose (for example, a design model for customer management), server 1 sends a prompt to the AI ​​agent that includes an instruction to implement it by referring to the design reference layer 134, which has been input, for example via terminal 2. At this time, since the common code layer 132 is already incorporated into the design reference layer 134, the AI ​​agent can generate source code that integrates the underlying common functions with business-specific designs and settings simply by referring to the design reference layer 134, without separately referring to the common code layer 132.

[0064] Furthermore, when building a new application 150, server 1 may, for example, specify an implementation pattern held in the already constructed design reference layer 134 as the design model for the new application via terminal 2. Then, it may instruct the AI ​​agent to reproduce the functions and design of the design reference layer 134. In this way, by using an instruction structure that directly references only the design reference layer 134, the amount of prompt description can be minimized, the inference load on the AI ​​agent can be reduced, and the quality and consistency of the generated application can be maintained at a high level. This is because complex dependencies can be resolved in advance on the infrastructure side (within the design reference layer 134), allowing for the rapid provision of a highly structured application 150 even when instructions are given via terminal 2 by a non-engineer.

[0065] Furthermore, an example of generating an upload screen for application 150 in this embodiment is shown. In this system, when generating an upload screen (file import screen) for extracting and saving data from an unstructured data file (e.g., an Excel® file), the prompt may include detailed information defining the data extraction rules and the structure of the storage destination. Specifically, firstly, the read logic 142 may include instructions regarding the database structure (table definition) where the extracted data will be stored, which may include a mapping definition of which master corresponds to each extracted column, and a definition of the relationships between each data. It may also include instructions for the system to supplement any missing data items. This makes it possible to generate consistent structured data even from incomplete unstructured data. Secondly, the read position 144 may include the specification of the starting row and column in the unstructured data file as information that identifies the data extraction range, as well as the extraction termination condition. This ensures that the data extraction range is selected more accurately. Thirdly, the reference to the design reference layer 134 may include a reference instruction to existing assets to ensure consistency in the design and functionality of the generated screen. Specifically, the design reference layer 134 may include a description that specifies a screen from a particular existing project and instructs the system to mimic and reproduce the functionality and design of that screen. This ensures that the proven UI / UX configuration held in the design reference layer 134 is accurately inherited by the newly generated upload screen.

[0066] Figure 5 is a diagram showing the processing flow when generating an application 150 according to one embodiment of the present invention. Figure 5 is a flowchart showing the processing procedure when building an application 150 using an AI agent in this system. In this flow, the server 1 dynamically configures prompts that combine assets of the common infrastructure 130 based on individual reading requirements sent from the user terminal 2b, thereby causing the AI ​​agent to generate an operational application. The details of each step will be described below.

[0067] Step S110 represents the information reception step. The communication unit 160 receives information from terminal 2 (for example, developer terminal 2a or user terminal 2b) regarding the read logic 142 and the read position 144 for an unstructured data file. Here, the read logic 142 includes instruction information regarding the conversion rules of the extracted data, and the read position 144 includes coordinate information indicating the location of the data in an unstructured spreadsheet, etc. In other embodiments, it is not necessary to receive information regarding both the read logic 142 and the read position 144, or either one thereof. In particular, if information regarding both is not received, this flow may start from the prompt transmission step in step 120.

[0068] Step S120 represents the prompt transmission process. The application generation unit 112 generates a prompt that includes the information received in step S110 and a reference instruction to the common infrastructure 130 held in the storage unit 120. Here, the reference instruction to the common infrastructure 130 instructs the AI ​​agent to use existing program assets and design models defined in the common code layer 132 and the design reference layer 134 during generation. The communication unit 160 transmits the generated prompt to the AI ​​agent server 3.

[0069] Step S130 represents the application reception process. The communication unit 160 receives the application 150 generated by the AI ​​agent server 3 in response to the prompt. The received data includes source code, execution scripts, or definition files for configuring the generated application 150. The application generation unit 112 stores this received data in the area of ​​the generated application 150 in the storage unit 120 and deploys it to an executable state in the application execution unit 114.

[0070] Figure 6 is a diagram showing the execution flow of a generated application 150 according to one embodiment of the present invention. Figure 6 is a flowchart showing the processing procedure for executing the generated application 150 and correcting settings (self-setting) in response to changes in the unstructured data file. This flow defines a series of steps to optimize the application 150 to a production-ready level by immediately deploying the constructed application 150 to the operational environment, performing verification using actual data, and dynamically adjusting parameters such as the reading position by the developers or users themselves. The details of each step are described below.

[0071] Step S210 describes the display process. The application execution unit 114 starts the generated application 150 and displays an operation screen on the display unit 240 of terminal 2 via the network N and the communication unit 160. This operation screen includes an upload interface for inputting unstructured data files into the system.

[0072] Step S220 indicates the upload acceptance process. The communication unit 160 accepts the upload of unstructured data files used in practical work from the terminal 2. The uploaded files are temporarily held in the storage unit 120, etc., via the control unit 110.

[0073] Step S230 describes the conversion process. Based on the retained read definition data 140 (read logic 142 and read position 144), the generating application 150 performs structuring processing, including data definition and table conversion, for the uploaded unstructured data file.

[0074] Step S240 represents the preview process. The generation application 150 generates a preview screen that includes the processing results from step S230 (for example, a preview of the extracted data or a dashboard display) and displays it on the display unit 240 of terminal 2. This allows the user to confirm whether the data has been read as intended.

[0075] Step S250 represents the correction and parameter update process. The control unit 110 accepts correction input from the terminal 2 as needed. For example, if the reading position is shifted due to a change in the format of an unstructured data file, the user gives a correction instruction via the screen on the display unit 240. Based on this instruction, the control unit 110 can rewrite parameters such as the reading position 144 and reflect the updated settings in the subsequent execution process.

[0076] Based on the above explanation, the various aspects of this embodiment will be described below.

[0077] [Process for generating application 150] In an application generation method according to one embodiment of the present invention, a computer (e.g., server 1) generates an application 150 that extracts and visualizes data from an unstructured data file. First, server 1 refers to a common base 130 held in storage unit 120. The common base 130 includes both or either a common code layer 132 that includes definitions of general-purpose functions and a design reference layer 134 that includes implementation patterns suitable for specific business purposes. The control unit 110 of server 1 sends a prompt to the AI ​​agent via the communication unit 160 to generate the generated application 150. This prompt includes a read logic 142 that defines rules (calculations, string manipulation, type conversion, etc.) for data extraction from an unstructured data file. Furthermore, the prompt includes an instruction to refer to at least one of the common code layer 132 and the design reference layer 134 of the common base 130 in generating the application 150. This enables the AI ​​agent to generate program code based on pre-verified reliable common code and successful design patterns while satisfying individual extraction requirements. As a result, while conventional DX (Digital Transformation) often imposes the need for unified input formats and data cleansing on the ground, leading to high workloads and frequent failures in implementation, and traditional AI-generated code has issues such as inconsistent quality and structural inconsistencies in generated output, as well as development periods of several weeks, this embodiment addresses these challenges by maintaining existing business flows and file formats (unstructured data) on the ground. The AI ​​agent absorbs these differences and structures the data, enabling the organization-wide information sharing (DX) to be realized ahead of schedule without imposing new burdens on the ground. Furthermore, it offers the remarkable benefit of enabling robust applications capable of withstanding actual operation to be completed by non-engineers within a single day (e.g., within a few hours) from the upload of unstructured data files. This allows management to make rapid business decisions based on dashboards that reflect the real-time situation on the ground.Furthermore, if the visualization results of this embodiment identify data inconsistencies, low update frequency, or significant stagnation in operational efficiency at a specific site, the visualization results can be used as evidence to move on to individual DX decisions for that site (such as a fundamental review of business processes or the introduction of dedicated tools). In other words, this embodiment also functions as a diagnostic infrastructure to identify areas where DX should be prioritized from the perspective of overall optimization.

[0078] [Modes of Dynamic Selection in Common Infrastructure Reference Instructions] In the above method, the reference instruction for the common infrastructure 130 may be selected from referencing only the common code layer 132, referencing only the design reference layer 134, or referencing both the common code layer 132 and the design reference layer 134, depending on the purpose of generating the application 150. In this embodiment, the configuration in which the reference instruction to the common infrastructure 130 is selected from the common code layer 132, the design reference layer 134, or a combination thereof plays an important role in improving development efficiency and ensuring the quality of the application 150. Examples of specific modes according to the generation purpose are described in detail below.

[0079] (1) Mode that refers only to the common code layer 132: This mode is selected primarily when the purpose is to reuse code. For example, it is suitable when reusing a general-purpose implementation that is functionally complex, such as a data display table or advanced validation checks, which has a large number of lines of code, while building the UI / UX design independently based on new requirements. It is also suitable for product-oriented generation purposes where the goal is to complete the setup completely without coding by applying only the common code when the product is highly packaged as SaaS (Software as a Service).

[0080] (2) Aspects that refer only to design reference layer 134: This aspect is selected primarily when the purpose is to reuse the design. For example, it is suitable when you want to reproduce the data structure of a data upload function in a specific business domain or the optimal UI / UX design, but want to implement the internal logic from scratch using your own technology stack. It is suitable for generation purposes where you want to inherit best practices (types) of the design while ensuring flexibility in implementation.

[0081] (3) A mode that references both the common code layer 132 and the design reference layer 134: This mode is selected when code reuse and design reuse are performed simultaneously, and the application 150 is completed with minimal new implementation. By making the already created design immediately reusable and incorporating frequently occurring implementation patterns as common code, it is ideal for generation purposes that aim to minimize the development burden using the AI ​​agent and deploy in an extremely short period of time.

[0082] The configuration may allow the developer or user to explicitly select which of the three patterns described above is used to generate application 150 via the operation screen of terminal 2. Alternatively, the server 1 may automatically determine and dynamically select the optimal reference pattern based on the timing of application 150 generation, the type of unstructured data file to be targeted, or past generation history. In other words, in this embodiment, the method for determining the reference pattern of each layer included in the common infrastructure 130 can encompass a wide range of forms, from those that directly reflect the user's intent to those that the system optimizes according to the situation.

[0083] In its most basic form, the developer or user may explicitly select a reference pattern via the terminal 2 on the development settings screen displayed on the display unit 240 of the terminal 2. This simply means that the developer or user may, when creating a prompt, include both or either the common code layer 132 and the design reference layer 134 in the prompt via the terminal 2. Alternatively, options such as "Prioritize common functions (code reuse)," "Comply with design model (design reuse)," and "Full scratch (focus on new implementation)" may be placed on the screen, or toggle switches or sliders to enable / disable each layer, allowing the scope of reference to be determined before application generation. This ensures that the specific intentions of the developer or user (e.g., "I want the same user experience as the existing core system") are reliably reflected.

[0084] Furthermore, automatic selection based on input data and context (context-sensitive selection) may be performed. For example, Server 1 (control unit 110) may automatically determine a reference pattern based on the generation timing and input information of Application 150. Specifically, it analyzes the file name, extension, internal header items, or similarity of the data structure of the uploaded unstructured data file and matches it with past successful generation examples (design reference layer 134). If the file contains many features related to, for example, "sales management," it may prioritize referencing the design model for "sales dashboard" from the design reference layer 134. On the other hand, if only general-purpose data cleaning is required, priority may be given to reusing functions by the common code layer 132.

[0085] In a more advanced form, dynamic optimization (heuristic selection) based on past learning and generation history may be performed. Past generation history and the frequency of subsequent user modifications (amount of modification in step S250) may be learned to determine the optimal reference pattern. For example, if there is a history in a particular business domain where applying only the common code layer 132 resulted in a large amount of subsequent modifications, while using the design reference layer 134 in combination resulted in a small amount of modifications, then Server 1 may prioritize selecting the latter pattern for similar generation purposes in the future. This allows a generation pattern that requires less rework to be autonomously established the more the system is used.

[0086] Furthermore, multi-objective optimization and autonomous hierarchical selection (autonomous meta-learning) by AI may be performed. It is also conceivable that AI agents such as large-scale language models (LLMs) or multi-agent systems predict the complexity, operational costs, and performance (e.g., consumption of computing resources) of the generated application, and autonomously determine the configuration that is most maintainable and satisfies user requirements from among multiple reference patterns. In this case, the AI ​​agent can not only simply select existing layers, but also extract elements from the vast amount of code fragments and design templates existing within the common infrastructure 130 at the optimal granularity for the generation purpose at that time, and dynamically construct and reference virtual intermediate layers.

[0087] Thus, the selection of reference patterns may be configured to continuously optimize the generated application 150 based on user feedback and autonomous AI decisions, in addition to selections made by current developers or users via terminal 2 or fixed rule-based selections. Furthermore, by separating the targets for reuse into layers of functional implementation code and purpose-appropriate designs, and selecting or combining these according to the generation purpose, it becomes possible to gradually increase the reused parts and continuously improve application development by the AI ​​agent. As a result, the AI ​​agent can autonomously build high-quality applications that can withstand actual operation based on existing trusted code assets and successful design patterns.

[0088] [Execution and Preview Display of Application 150] When the generated application 150 is executed, it displays multiple screens on the display unit 240 of terminal 2. Specifically, the generated application 150 may display an upload screen that accepts input of an unstructured data file, and a preview screen that includes the data structure defined based on the reading results and a display of the converted table format. That is, the generated application 150 has an upload interface that prompts the user to select an unstructured data file immediately after startup. The preview screen may not only display the data, but may also display the definition of the data structure interpreted by the AI ​​(column names and types) alongside the actual data table. By referring to this preview screen, the user can visually confirm whether the unstructured source data has been correctly converted into structured data.

[0089] [Correction process using self-setting function] Furthermore, the generated application 150 may, at runtime, receive correction instructions from the developer or user via terminal 2 through the preview screen regarding at least one of the reading logic 142 or the reading position 144 that specifies the range of data extraction on an unstructured data file (such as matrix coordinates of a spreadsheet). When such correction is received, the control unit 110 may directly rewrite the parameters of the reading settings in the reading definition data 140 (for example, numerical values ​​such as the starting row and column of a cell) without rebuilding (regenerating) the entire application. For example, operators (such as sliders or input forms) for directly adjusting the reading position (matrix range, etc.) may be placed on this screen, and when a user operates these via terminal 2b, only the setting parameters in memory may be immediately updated without involving development processes such as regenerating the entire program or recompiling the source code in the backend. This configuration allows for immediate optimization of operation through user configuration alone, without requiring a regeneration process by the AI ​​agent, in response to minor changes in file format or shifts in reading position, enabling continuous production operation. In other words, with conventional template methods, even minor changes to data reading position or aggregation logic required regenerating, compiling, and deploying the source code. This made it difficult for non-engineer field personnel to build a real-time visualization environment through trial and error. In contrast, this embodiment combines flexible code generation by the AI ​​agent utilizing the common platform 130 with dynamic parameter update functionality at runtime, enabling immediate setting adjustments (self-setting) without rebuilding the application, dramatically improving operational continuity and convenience. This behavior can be objectively confirmed externally through the immediate linkage of operations and displays on the user interface. In other embodiments, if the server 1 receives such modifications, the control unit 110 may rebuild (regenerate) the entire application. For example, when a developer is generating an initial version of application 150, they may specify the modification in the prompt via terminal 2a and send it to the AI ​​agent server 3.This allows developers to restart the process early if an unintended application 150 is being generated, and as a result, the generation of the final application 150 can be completed more quickly.

[0090] [Program Structure] The program according to this embodiment causes a computer to execute each of the above-described steps. Here, the computer executing the program in this specification may be a single physical device (e.g., Server 1), or it may not be limited thereto, but may encompass an entire system in which multiple information processing devices interconnected via a network N work in cooperation. Specifically, the computer may mainly consist of Server 1, which controls the generation and execution of applications, while utilizing at least some or all of the functions of an AI agent server 3 that performs code generation processing in response to prompts, and a terminal 2 that provides a user interface and is responsible for uploading data and displaying a preview screen, as a logically integrated computing resource. For example, even if some of the calculation processing in each of the above-described steps (S110 to S130 and S210 to S250) is executed on the AI ​​agent server 3 side and data input / output control is shared and executed on the terminal 2 side, a series of instructions that control these distributed processes are included in the program according to this embodiment. In this way, the computer that executes the program may be configured as a virtual processing platform in which multiple physically separated hardware devices (Server 1, Terminal 2, and AI agent server 3) work in cooperation to realize a series of workflows. This ensures that the technical scope of the present invention as a program extends equally across various implementation forms in cloud computing environments.

[0091] [Device Configuration] The device (e.g., Server 1) that implements each of the steps described above comprises a control unit 110, a storage unit 120, and a communication unit 160. Here, "device" in this specification is not limited to a computer housed in a single physical enclosure, but also includes a virtual group of servers configured by dynamically allocating computing resources on a network, or an information processing system in which multiple geographically distributed devices communicate with each other to function as a single device. The control unit 110 is composed of one or more processors (CPU, GPU, or application-specific integrated circuit, etc.) and operates as various functional blocks (application generation unit 112, application execution unit 114, etc.) by executing a program. This control unit 110 is not limited to being located within Server 1, and some of its computing functions may be distributed and located on the AI ​​agent server 3 or terminal 2 side. The storage unit 120 may be a physical memory device, or a distributed database structure such as a storage area network (SAN) or cloud storage connected via a network. The common base 130, read definition data 140, and generated application 150 described above are logically stored in this storage unit 120. These do not need to be stored together in a single storage medium; they may be distributed and managed across multiple storage areas (e.g., sharding) for the purpose of optimizing confidentiality and access speed. The communication unit 160 is a communication interface compatible with various wired or wireless communication protocols, and establishes an encrypted and secure data transmission path between the server 1, terminal 2, and AI agent server 3. In this way, the "device" of the present invention can be configured as a platform in which multiple computing nodes and storage nodes are functionally coupled, without being bound by physical boundaries. As a result, the technical scope of the device of the present invention is encompassed regardless of whether the overall system configuration is an on-premises environment, a hybrid cloud, or a multi-cloud environment.

[0092] [Implementation and Operational Forms] The system can take the following forms from the implementation to the operation of the generated application 150. First, during implementation, Server 1 generates Application 150 based on an unstructured data file provided by the user, for example, by having a developer determine the initial reading logic 142 and, if necessary, the reading position 144 via Terminal 2a. During operation after generation, if there is a discrepancy in the starting position of the data in the file, the developer or user corrects the reading settings via the user interface on the preview screen of Terminal 2 (for example, an input form or slider for specifying the cell position). In this case, Server 1 may immediately reflect the correction results on the preview screen by dynamically rewriting only the setting parameters in the application without re-requesting the AI ​​agent or recompiling the code, or in the initial stage, the developer may instruct the AI ​​agent to regenerate Application 150 via Terminal 2a. Next, as another form of implementation, the user may input a description of the file structure and extraction rules on the screen of Terminal 2b, and Server 1 may automatically configure prompts to send to the AI ​​agent based on this input. This makes it possible to construct a reading logic 142 that takes into account the intentions of non-engineer users. Furthermore, in another operational scenario, if the data extraction rules themselves need to be changed, the user may instruct the terminal 2b to change the reading logic 142 from its screen. Upon receiving this instruction, the server 1 may reconfigure the prompt and rebuild (regenerate) the program code of the application 150 itself via the AI ​​agent.

[0093] In this system, when a developer or user creates a new application 150, they may reuse the configuration information of other applications 150 previously created and stored in the storage unit 120 of the server 1 via terminal 2 to build the new application 150. Specifically, the developer or user accesses the management screen via terminal 2 and selects a target for reuse from the list of existing projects accumulated in the storage unit 120. In response to this operation, the control unit 110 of the server 1 may reuse the configuration data of prompts associated with the selected existing project, the definition of the referenced common infrastructure 130, and setting information such as operating parameters, and deploy them as the initial settings for the new project. Based on the reused setting information, the developer or user modifies the origin (connection source information) of the data source, the application name, or unique information such as individual extraction coordinates via terminal 2 to ensure consistency with the new unstructured data file. This reuse operation by the developer or user makes it possible to inherit previously optimized extraction know-how (reading logic 142, etc.) when deploying the application to multiple sites or departments with similar data structures. This allows even non-engineers to quickly build 150 applications that maintain a certain level of quality by leveraging existing successful patterns.

[0094] As described above, the application generation method in this embodiment is a method in which a computer (e.g., Server 1) generates an application 150 that extracts and visualizes data from an unstructured data file, wherein the computer refers to a common infrastructure 130 that holds either or both a common code layer 132 containing definitions of general functions and a design reference layer 134 containing implementation patterns suitable for specific business purposes, and sends a prompt to an AI agent to generate the application 150, wherein the prompt includes read logic 142 defining rules for data extraction from the unstructured data file and an instruction to refer to at least one of the common code layer 132 and the design reference layer 134 of the common infrastructure 130 in generating the application 150. This makes it possible to quickly and easily generate an application with a certain level of quality guaranteed.

[0095] Furthermore, this embodiment offers the following advantages. Firstly, in application generation by the AI ​​agent, a hierarchical common infrastructure 130 is referenced, which includes a general-purpose common code layer 132 and a design reference layer 134 that is a specific implementation sample. This ensures the quality and consistency of the code generated by the AI ​​agent, making it possible to build sophisticated applications tailored to individual business requirements in a short amount of time.

[0096] Secondly, the system employs a configuration that separates and manages the read logic 142, which defines the data conversion procedure, and the read position 144, which specifies the location of the data. This allows the system to adapt immediately to minor changes in the layout of unstructured data files used in the field, simply by the user modifying parameters on the user interface, without requiring redevelopment by engineers.

[0097] Thirdly, the prototype in the development stage and the production environment are seamlessly integrated. According to this embodiment, the environment in which users have verified and adjusted using actual data during the prototyping process can be put into operation as a fully operational product. This significantly reduces the lead time to implementation and enables rapid real-time visualization of unstructured data.

[0098] Overall, this embodiment solves the challenges of securing engineering resources and adapting to non-standard operations in conventional system development, which have been major obstacles in the past. It provides a data visualization platform that can be continuously operated even by non-engineers with minimal investment and without burdening the field. In particular, it allows the system to flexibly adapt to non-standard data structures that differ from site to site without changing the input rules at the site, and minimizes the implementation lead time by replacing the development process with a work process.

[0099] (Modifications) The embodiments described above are illustrative examples for explaining the technical concept of the present invention, and the present invention is not limited thereto. Various modifications are possible without departing from the spirit of the present invention, as follows.

[0100] [Modifications of Device Configuration and Arrangement (1) Integration and Distribution of Devices] In the above embodiment, Server 1, Terminal 2, and AI Agent Server 3 were described as separate devices, but all or part of these functions may be aggregated on the same physical or logical computing resources. For example, Server 1 may be equipped with an inference engine as an AI agent (such as a local LLM or multi-agent), and may be configured to generate applications without going through an external network. Alternatively, Terminal 2 may encompass the functions of Server 1 and operate as a standalone device.

[0101] [Modifications of device configuration and arrangement (2) Application to edge computing] A distributed arrangement may be adopted in which some processing, such as data extraction processing from unstructured data files and preview display generation processing, is performed on terminal 2 (edge ​​device), and only the management of the common platform 130 and prompt configuration are performed on server 1.

[0102] [Variations regarding functions, data, and files (1) Diversity of unstructured data files] The unstructured data files covered are not limited to spreadsheet format. Any unstructured data file format may be covered, including CSV, PDF, text files, image files (through OCR processing), and even email bodies, chat logs, and text conversion results of audio data.

[0103] [Variations regarding functions, data, and files (2) Hierarchical extension of the common platform 130] The common platform 130 is not limited to two layers: the common code layer 132 and the design reference layer 134. For example, it may have a hierarchical structure of three or more layers, such as an "industry-specific layer" that is specialized for a particular industry (finance, manufacturing, healthcare, etc.) or a "compliance layer" that defines compliance with specific laws and regulations.

[0104] [Modifications regarding functions, data, and files (3) Dynamic generation of read definition data 140] The read logic 142 and read position 144 may be configured such that the server 1 analyzes the file structure and automatically suggests (recommends) initial values, in addition to being specified by the developer or user via terminal 2. The configuration may also include sharing or templating the read definition data once it has been generated among multiple similar unstructured data files for reuse.

[0105] [Variations regarding generation and execution behavior (1) Sequential prompt configuration (multi-turn generation)] The generation of application 150 does not need to be completed with a single prompt transmission. A stepwise generation process may be adopted in which multiple interactions (multi-turn conversion) with the AI ​​agent are performed to first generate a skeleton (framework), and then each layer of the common infrastructure 130 is sequentially applied to flesh it out.

[0106] [Variations regarding generation and execution behavior (2) Form of generated application 150] The generated application 150 is not limited to a web application. It may also be a native application (for Windows® / Mac®), a mobile app (for iOS® / Android®), an add-on that runs on an existing ERP or CRM system, or output as a scenario for a workflow automation tool (such as iPaaS).

[0107] [Modifications regarding generation and execution operations (3) Automation of self-setting] In the execution phase, if a change in the layout of an unstructured data file is detected, the server 1 may autonomously correct the reading position 144 based on past correction patterns without user intervention, thus enabling an automated self-setting configuration.

[0108] [Variations regarding generation and execution operation (4) Multilingual and multi-domain support] In generating the read logic 142, it may be configured to support prompt configurations in multiple languages ​​as well as Japanese, and to integrate files in different formats at each global location into a unified data structure using the common platform 130.

[0109] [Variations regarding generation and execution operation (5) Format saving and pull-down selection function] The upload screen may be provided with a selection section (pull-down menu, etc.) for saving and reusing previously confirmed read definition data 140 as a file format (template). When a user selects a specific format and uploads a new file, Server 1 skips the new analysis process by the AI ​​agent and immediately applies the known read logic 142 and read position 144. Specifically, the upload screen (not shown) displayed on the display unit 240 of Terminal 2 may be provided with an interface for uploading unstructured data files, along with a selection pull-down menu for specifying a file format (read definition template, file read settings, etc.) that is suitable for the file being uploaded. When a user uploads a file via Terminal 2b with a specific file format selected using the pull-down menu, the application execution unit 114 of Server 1 applies the read definition data 140 (read logic 142 and read position 144) corresponding to the selected format. This enables high-precision and high-speed data extraction based on a known layout without waiting for new analysis by the AI ​​agent. Therefore, this improves processing speed in routine tasks and stabilizes extraction accuracy. The selection dropdown may list formats previously created by the user or other users, as well as general-purpose formats provided by the system. If a suitable format does not exist, a mode to generate a new reading definition may be selected.

[0110] [Modifications regarding generation and execution behavior (6) Multi-faceted visualization function via tab switching] The display screen of the generation application 150 may be provided with multiple tabs for switching the data display format and analysis perspective. For example, a "table tab" that structures and displays the contents of an uploaded unstructured data file in a list, a "dashboard tab" that visually represents extracted numerical data with graphs, charts, etc., and a "settings tab" for checking and correcting the data reading range and logic may be dynamically generated. In particular, the "table tab" may include the data definition defined by the reading logic, the database and data for each item based on the said data definition, and the result of converting the unstructured data file into a table, and may be displayed as the initial screen of the preview screen, and the table conversion result may be equipped with a visual confirmation function such as coloring to indicate the presence or absence of data or inconsistencies. By the user selecting (clicking, etc.) these tabs via terminal 2b, server 1 can provide the optimal UI / UX via different rendering engines, even though it is based on the same data source, and support the confirmation of the success or failure of the upload and multi-faceted data analysis.

[0111] [Variations regarding generation and execution operation (7) Batch application and difference extraction function] This system may have a function to apply a read definition once set to multiple unstructured data files with a common layout, not just a single file. Furthermore, even if there are minor differences in the layout between files, a base setting may be selected, and only specific matrix ranges may be individually modified using self-setting, and these may be saved as a different derived format.

[0112] [Other technical variations (1) Utilization of blockchain, etc.] Blockchain technology may be used for the update history of the common platform 130 and for version control of the generated applications, ensuring the authenticity of design assets and the immutability of change history.

[0113] [Other technical variations (2) Self-propagation of the base through a feedback loop] The server 1 periodically analyzes the user modifications made in step S250 and automatically promotes (feeds back) highly versatile modification patterns to the common code layer 132 or design reference layer 134, thereby enabling the common base 130 to evolve on its own.

[0114] Furthermore, the present invention may be realized as described below.

[0115] [Item 1] A method for a computer to generate an application for extracting and visualizing data from an unstructured data file, the computer referring to a common infrastructure that holds either or both a common code layer containing definitions of general functions and a design reference layer containing implementation patterns suited to specific business purposes, and sending a prompt to an AI agent to generate the application, the prompt including read logic that defines rules for extracting data from the unstructured data file, and instructions that refer to at least one of the common code layer and the design reference layer of the common infrastructure in generating the application.

[0116] [Item 2] The method according to Item 1, wherein the instruction to reference the common infrastructure in the prompt is, depending on the purpose of generating the application, a reference to the common code layer only, a reference to the design reference layer only, or a reference to both the common code layer and the design reference layer.

[0117] [Item 3] The method according to Item 1, wherein the generated application, at runtime, causes the terminal's display unit to show a preview screen including an upload screen for the unstructured data file, a data structure defined based on the reading results of the unstructured data file, and a display of the converted table format.

[0118] [Item 4] The generated application, at runtime, receives a modification from the terminal via the preview screen to either the reading logic or the reading position that identifies the matrix range for data extraction on the unstructured data file, and if such modification is received, rewrites the parameters of the reading settings without rebuilding the entire application, according to the method of Item 3.

[0119] [Item 5] A program that causes a computer to perform any of the methods described in items 1 through 4.

[0120] [Item 6] A device for generating an application for extracting and visualizing data from an unstructured data file, the device comprising a control unit, a storage unit, and a communication unit, wherein the storage unit stores a common base holding either or both a common code layer including definitions of general functions and a design reference layer including implementation patterns suitable for specific business purposes, the control unit refers to the common base and sends a prompt to an AI agent via the communication unit for generating the application, the prompt including read logic defining rules for data extraction from the unstructured data file and instructions to refer to at least one of the common code layer and the design reference layer of the common base in generating the application.

[0121] [Item 7] A method for a computer to generate an application for extracting and visualizing data from an unstructured data file, the computer referencing a common infrastructure that holds either or both of a common code layer containing program code defining reusable general-purpose functions and a design reference layer containing implementation examples defining design patterns for data structures and user interfaces suited to specific business purposes, and sending a prompt to an AI agent to generate the application, the prompt including read logic that defines rules for extracting data from the unstructured data file and instructions that in generating the application refer to at least one of the common code layer and the design reference layer of the common infrastructure, the AI ​​agent, upon receiving the prompt, generates source code in a form that calls the program code contained in the common code layer, with the implementation examples held in the design reference layer serving as a design reference model.

[0122] [Item 8] The method of Item 1 or 7, wherein the instruction to reference the common infrastructure in the prompt is dynamically selected from referencing only the common code layer, referencing only the design reference layer, or referencing both the common code layer and the design reference layer, depending on the purpose of generating the application, or is determined based on a selection made by the developer or user via a terminal.

[0123] [Item 9] The method according to Item 7, wherein the generated application causes the terminal's display unit to show, at runtime, an upload screen for the unstructured data file, and a preview screen including a data structure defined based on the reading results of the unstructured data file, and a table format display converted according to the data structure.

[0124] [Item 10] The method according to Item 3 or 9, wherein the generated application, at runtime, receives at least one of the instructions to modify the reading logic or the instructions to modify the reading position that specifies the matrix range for data extraction on the unstructured data file via the preview screen, and when the instruction to modify the reading position is received, the parameters of the reading settings within the application are rewritten and the display on the preview screen is updated without rebuilding the entire application.

[0125] [Item 11] A program that causes a computer to perform any of the methods described in items 7 through 10.

[0126] [Item 12] A device for generating an application for extracting and visualizing data from an unstructured data file, the device comprising a control unit, a storage unit, and a communication unit, wherein the storage unit stores a common base holding either or both of a common code layer containing program code defining reusable general-purpose functions and a design reference layer containing implementation examples defining design patterns for data structures and user interfaces suitable for specific business purposes, the control unit refers to the common base and sends a prompt to an AI agent via the communication unit for generating the application, the prompt including read logic defining rules for data extraction from the unstructured data file and instructions to refer to at least one of the common code layer and the design reference layer of the common base in generating the application, and the AI ​​agent generates source code in a form that calls the program code defined in the common code layer, using the implementation examples held in the design reference layer as a design reference model.

[0127] [Item 13] The method according to Item 3 or 9, wherein the display unit of the terminal displays a selection unit for selecting a specific format from a set of pre-configured file formats when uploading the unstructured data file, and when a specific format is selected via the selection unit, the application applies both or either the reading logic and the reading position associated with that format to the unstructured data file to generate the preview screen.

[0128] [Item 14] The apparatus according to Item 6 or 12, wherein the storage unit holds multiple previously generated read logic and / or read positions as file formats, and the control unit, upon receiving an upload of the unstructured data file from the terminal with the specification of a specific file format, reads the specified file format from the storage unit and applies it to the application.

[0129] [Item 15] The generated application is the method according to Item 3 or 9, wherein the terminal's display unit displays a first display area that displays the data extracted from the unstructured data file in a tabular format and a second display area that displays the data visually as a diagram or chart, and these two areas can be switched between using tabs.

[0130] [Item 16] The method according to Item 15, wherein the first display area includes the preview screen and an input interface for modifying at least one of the reading logic and the reading position, on the same screen.

[0131] [Item 17] The method according to Item 1 or 7, wherein the computer applies the read logic set for one unstructured data file to a plurality of other unstructured data files in a batch, and integrates the data extracted from each file into a common data structure.

[0132] [Item 18] The method according to Item 3 or 9, wherein the generated application displays a preview screen on the terminal's display unit at runtime that allows comparison between the reading results of the unstructured data file and the mapping results to a table structure based on the design reference layer, receives correction instructions for the mapping results via the preview screen, and updates at least one of the reading logic or the reading position based on the correction instructions.

[0133] [Item 19] A system for generating an application that extracts and visualizes data from an unstructured data file, comprising: a storage unit that stores a common infrastructure that holds either or both a common code layer containing definitions of general functions and a design reference layer containing implementation patterns suitable for specific business purposes; and a control unit that sends a prompt to an AI agent for generating the application, wherein the prompt includes read logic that defines rules for data extraction from the unstructured data file and an instruction that in generating the application, references at least one of the common code layer and the design reference layer of the common infrastructure.

[0134] [Item 20] A device for generating an application that extracts and visualizes data from an unstructured data file, comprising: a storage unit that stores a common base that holds either or both a common code layer that includes definitions of general functions and a design reference layer that includes implementation patterns suitable for specific business purposes; and a control unit that obtains source code constituting the application by sending a prompt to an AI agent that includes a reference instruction to the common base, wherein the application outputs a preview screen based on the reading results of the unstructured data file to the display unit of a terminal, and when a correction of the reading position indicating the coordinates of data extraction in the unstructured data file is received via the preview screen, the device updates the parameters of the reading position and rewrites the preview screen without regenerating the application.

[0135] [Item 21] The apparatus according to Item 20, wherein the prompt includes read logic that includes a logical procedure for extracting data from the unstructured data file, and the AI ​​agent generates the source code in a form that calls the program code contained in the common code layer, using the implementation patterns contained in the design reference layer as a design model.

[0136] [Item 22] The apparatus described in Item 20 or 21, wherein the reference instruction to the common platform is, depending on the purpose of generating the application, a reference to the common code layer only, a reference to the design reference layer only, or a reference to both the common code layer and the design reference layer.

[0137] [Item 23] The apparatus according to Item 21, wherein the application, upon receiving a change instruction for the read logic, reconfigures the prompt via the control unit and regenerates the application. [Industrial applicability]

[0138] This invention can be widely used as a data utilization platform that converts unstructured data files such as spreadsheets and various forms, which are routinely used in a wide range of industrial fields such as manufacturing, retail, finance, and healthcare, into visualization applications that can be immediately executed via an AI agent. It is particularly highly practical in low-code / no-code platforms that support rapid DX (Digital Transformation) driven by non-engineers, or in information processing service industries such as SaaS. [Explanation of symbols]

[0139] 1 server 110 Control Unit 112 Application Generation Section 114 Application Execution Unit 120 Storage section 130 Common Platform 132 Common Code Layer 134 Design Reference Layers 140 Read definition data 142 Reading Logic 144 Reading position 150 generated applications 160 Communications Department 2 terminals 2a Developer terminal 2b User terminal 210 Control Unit 220 Storage section 230 Communications Department 240 Display section 3 AI Agent Server N Internet

Claims

1. A method for a computer to generate an application that extracts and visualizes data from an unstructured data file, The computer references a common infrastructure that holds either or both a common code layer containing definitions of general functions and a design reference layer containing implementation patterns suitable for specific business purposes. The process includes sending a prompt to the AI ​​agent to generate the application, The aforementioned prompt will include: A reading logic that defines the rules for extracting data from the aforementioned unstructured data file, In generating the aforementioned application, an instruction is given to reference at least one of the common code layer and the design reference layer of the common infrastructure, The method, which includes the above.

2. The computer refers to the common infrastructure which holds both the common code layer and the design reference layer, The method according to claim 1, wherein the instruction to reference the common infrastructure in the prompt is, depending on the purpose of generating the application, a reference to the common code layer only, a reference to the design reference layer only, or a reference to both the common code layer and the design reference layer.

3. The generated application, at runtime, The aforementioned non-standard data file upload screen, and A preview screen is provided that includes a data structure defined based on the reading results of the aforementioned unstructured data file, and a display of the converted table format. The method according to claim 1, wherein the information is displayed on the display unit of the terminal.

4. The generated application, at runtime, The terminal accepts a modification from the terminal via the preview screen to either the reading logic or the reading position that identifies the matrix range for data extraction on the unstructured data file. The method according to claim 3, wherein, upon acceptance of the said modification, the parameters of the read settings are rewritten without rebuilding the entire application.

5. A program that causes a computer to perform the method described in any one of claims 1 to 4.

6. A device that generates an application for extracting and visualizing data from unstructured data files, The device comprises a control unit, a storage unit, and a communication unit. The memory unit stores a common infrastructure that holds either or both a common code layer containing definitions of general functions and a design reference layer containing implementation patterns suitable for specific business purposes. The control unit, referring to the common platform, sends a prompt to the AI ​​agent via the communication unit to cause it to generate the application. The aforementioned prompt will include: A reading logic that defines the rules for extracting data from the aforementioned unstructured data file, In generating the aforementioned application, an instruction is given to reference at least one of the common code layer and the design reference layer of the common infrastructure, The apparatus includes the aforementioned device.

Images