Method and system for work assistant using artificial intelligence agent and computer program for the same

Abstract

A method and system for assisting work using an AI agent and a computer program for the same are disclosed. A method for assisting work using an AI agent executed by a computing device may include: receiving template information that defines the format of a work result; a step in which an AI agent executed on the computing device acquires one or more of a screen image and text of the computing device containing the user's work content; a step of generating analysis information for a work using a feature value extracted from the template information and one or more of the screen image and text as input values ​​for a pre-trained large-scale language model; and a step of processing the analysis information into work result data corresponding to the template information.

Description

Technology Field

[0001] The embodiments relate to a method and system for assisting work using an AI agent and a computer program for the same. More specifically, the embodiments relate to a technology that automatically acquires images and / or text of a user's work screen through an AI agent, generates analysis information by applying feature values ​​extracted from user-specified template information and acquired work content to a large-scale language model, processes this into work result data in a template format, and enables it to be displayed, modified, and saved. Background Technology

[0002] Recently, AI agents based on artificial intelligence technology, particularly Large Language Models (LLMs), are being utilized in various work environments. By providing functions to automate or assist with numerous tasks such as natural language processing, document creation, and data analysis, these AI agents enable users to reduce their work time and maintain work quality at a certain level or higher.

[0003] However, conventional AI-based work assistance tools are limited to providing data in the form of answers when a user inputs a query through a specific web or app interface. When users perform actual work, they use word processors, spreadsheets, presentation programs, etc., but conventional AI agents are limited to operating in conjunction with some information from these applications and do not provide the ability to immediately acquire and analyze the content of work screens or input text generated from various applications.

[0004] Furthermore, various applications each have their own proprietary output data formats, and since there is no process to convert them into standardized output formats, users faced the inconvenience of having to manually organize and convert data formats according to their requirements every time, even when receiving responses from AI agents.

[0005] In other words, conventional AI agents mostly operate separately from the applications used by users for actual work, and even when information is obtained through the AI ​​agent, separate manual work or post-processing is often required to generate results tailored to specific formats (HWP, DOCX, XLSX, etc.), which has resulted in limitations in terms of connectivity with real-time workflows. Prior art literature

[0006] Korean Registered Patent No. 10-2392359 (Registered on April 26, 2022) The problem to be solved

[0007] The present invention aims to solve the aforementioned problems by providing a method and system for assisting work using an AI agent, and a computer program for the same, which automatically acquires images and / or text of a user's work screen through an AI agent, generates analysis information by applying feature values ​​extracted from user-specified template information and acquired work content to a large-scale language model, and provides this as work result data in a template format.

[0008] Furthermore, the present invention aims to provide a technology that enables standardized data collection and analysis in various work environments by having an AI agent recognize when a user activates a specific application and automatically acquire the corresponding screen image and / or text without separate manual operation, and by supplementing the text from the image through OCR processing.

[0009] In addition, the present invention aims to provide a technology that enables the result to be automatically completed in a form consistent with the template structure by applying feature values ​​extracted from template information to the output format settings of a large-scale language model and generating analysis information by combining content acquired from a work screen therewith.

[0010] In addition, the present invention aims to provide a technology that enables the application of a template selected by a user from among a plurality of template information prior to the generation of analysis information, thereby allowing the rapid reflection of a result format suitable for business purposes and requirements.

[0011] In addition, the present invention aims to provide a technology that can enhance the usability and completeness of results by displaying generated work result data on a screen through a user interface and allowing the user to directly modify and supplement it and then save it in a designated file format.

[0012] The problem to be solved by this specification is not limited to what is described above and can be extended to various matters that can be derived from the embodiments of the invention described below. means of solving the problem

[0013] A method for assisting work using an AI agent executed by a computing device according to one aspect of the present invention comprises: receiving template information that defines the format of a work result; an AI agent executed on the computing device acquiring one or more of a screen image and text of the computing device that includes the user's work content; generating analysis information for a work by using a feature value extracted from the template information and one or more of the screen image and text as input values ​​for a pre-trained large-scale language model; and processing the analysis information into work result data corresponding to the template information.

[0014] In one embodiment, the step of acquiring one or more of the screen image and text includes the step of automatically acquiring one or more of the screen image and text from the computing device in response to a preset application being activated on the computing device.

[0015] In one embodiment, the step of generating the analysis information includes: setting the output format of the large-scale language model using feature values ​​extracted from the template information; and inputting one or more of the screen image and text into the large-scale language model having the output format set.

[0016] A method for assisting work using an AI agent according to one embodiment further includes, prior to the step of generating the analysis information, a step of receiving a user's selection input regarding a template information to be used for the work result data among a plurality of template information pre-stored in the computing device or an external device capable of communicating with the computing device.

[0017] A method for assisting with work using an AI agent according to one embodiment further includes the step of displaying the work result data on the computing device through a user interface that allows the user to modify the work result data.

[0018] A computer program according to another aspect of the present invention is stored in a computer-readable recording medium to perform a task assistance method using an AI agent according to the embodiments described above, combined with hardware.

[0019] A work assistance system using an AI agent executed by a computing device according to another aspect of the present invention comprises: a template input unit configured to receive template information defining the format of the work result; a work content acquisition unit configured to acquire one or more of a screen image and text of the computing device including the user's work content in conjunction with the AI ​​agent; an analysis information generation unit configured to generate analysis information for the work by using a feature value extracted from the template information and one or more of the screen image and text as input values ​​for a pre-trained large-scale language model; and a result processing unit configured to process the analysis information into work result data corresponding to the template information.

[0020] In one embodiment, the work content acquisition unit is further configured to automatically acquire one or more of the screen image and text from the computing device in response to a preset application being activated on the computing device.

[0021] In one embodiment, the analysis information generation unit is further configured to set the output format of the large-scale language model using feature values ​​extracted from the template information, and to generate the analysis information by inputting one or more of the screen image and text into the large-scale language model having the output format set.

[0022] In one embodiment, the template input unit is further configured to receive a user's selection input regarding template information to be used for business result data among a plurality of template information stored in advance, before the analysis information is generated.

[0023] A work assistance system using an AI agent according to one embodiment further includes a user interface module configured to display the work result data on the computing device and provide it so that a user can modify it. Effects of the invention

[0024] According to one aspect of the present invention, images and / or text of a work screen are automatically acquired through an AI agent based on template information specified by a user, and analysis information is generated by applying feature values ​​extracted from the template information and acquired work content to a large-scale language model and then processing this into work result data in a template format, thereby providing the advantage of rapidly generating results in a standardized format without manual work.

[0025] In addition, according to one aspect of the present invention, when a specific application is activated, an AI agent automatically recognizes it and acquires screen images and / or text, thereby providing convenience so that the user does not have to perform a separate capture or copy procedure, and in the case of image-based content, there is an advantage of being able to supplement text data through OCR processing.

[0026] In addition, according to one aspect of the present invention, feature values ​​extracted from template information are applied to the output format settings of a large-scale language model, and analysis information is generated by combining content acquired from a work screen therewith, so there is an advantage that the result can be generated in a form that completely matches a predefined template structure.

[0027] In addition, according to one aspect of the present invention, since a template selected by the user among a plurality of templates can be applied prior to the generation of analysis information, there is an advantage of providing flexibility to generate customized results tailored to various business purposes and situations.

[0028] In addition, according to one aspect of the present invention, generated work result data is displayed through a user interface, and since the user can directly modify and supplement it and then save it in a designated file format, there is an advantage of increasing the completeness of the result and improving its usability for subsequent work.

[0029] In addition, according to one aspect of the present invention, since data acquisition, analysis, processing, display, and storage are integrated within a single system, there is an advantage of enabling stable and scalable work assistance regardless of the work environment or the user's skill level.

[0030] In addition, according to one aspect of the present invention, manpower consumption and time in repetitive document creation, report generation, and data standardization tasks can be reduced, thereby improving organizational productivity and increasing operational efficiency.

[0031] It should be understood that the effects of this specification are not limited to the matters described above and can be extended to various contents that can be derived from the detailed description of the embodiments of the invention below. Brief explanation of the drawing

[0032] FIG. 1 is a schematic diagram illustrating the operational structure of a business assistance system using an AI agent according to one embodiment of the present invention. FIG. 2 is a diagram illustrating the overall configuration of a work assistance system using an AI agent according to an embodiment of the present invention. FIG. 3 is a schematic diagram illustrating the hardware configuration of a business assistance system using an AI agent according to an embodiment of the present invention. FIG. 4 is a flowchart illustrating the process of receiving template information that defines the format of a business result through a template input unit according to an embodiment of the present invention. FIG. 5 is a flowchart illustrating the process of a work content acquisition unit according to an embodiment of the present invention acquiring a screen image and / or text of a computing device. FIG. 6 is a flowchart illustrating the process of a work content acquisition unit according to an embodiment of the present invention automatically acquiring a screen image and / or text in response to a preset application activation. FIG. 7 is a flowchart illustrating the process of an analysis information generation unit according to an embodiment of the present invention generating analysis information by inputting an image and / or text of a work screen into a large-scale language model in which the output format is set in advance. FIG. 8 is a flowchart illustrating the process of a template input unit according to an embodiment of the present invention receiving a template selected by a user among a plurality of template information before generating analysis information. FIG. 9 is a diagram illustrating the process of displaying business result data generated through a user interface module according to an embodiment of the present invention, and the user modifying and downloading it in a specified file format. FIG. 10 is a drawing illustrating an example of a template selection and automatic acquisition of work content UI according to an embodiment of the present invention. FIG. 11 is a drawing illustrating a result display, editing, and download UI configuration as an example of a user terminal screen according to an embodiment of the present invention. FIG. 12 is a flowchart illustrating the overall procedure of a task assistance method using an AI agent according to one embodiment of the present invention. Specific details for implementing the invention

[0033] In describing the embodiments of this specification, if it is determined that a detailed description of known configurations or functions could obscure the essence of the embodiments of this specification, such detailed description is omitted. Additionally, parts of the drawings unrelated to the description of the embodiments of this specification have been omitted, and similar parts are denoted by similar reference numerals.

[0034] In the embodiments of this specification, when a component is described as being "connected," "combined," or "joined" with another component, this may include not only a direct connection but also an indirect connection in which another component exists in between. Furthermore, when a component is described as "comprising" or "having" another component, this means that, unless specifically stated otherwise, it does not exclude the other component but may include additional components.

[0035] In the embodiments of this specification, terms such as first, second, etc. are used solely for the purpose of distinguishing one component from another component and do not limit the order or importance of the components unless specifically stated otherwise. Accordingly, within the scope of the embodiments of this specification, the first component in an embodiment may be referred to as the second component in another embodiment, and likewise, the second component in an embodiment may be referred to as the first component in another embodiment.

[0036] In the embodiments of this specification, distinct components are intended to clearly explain their respective features and do not imply that the components are necessarily separated. That is, multiple components may be integrated to form a single hardware or software unit, or a single component may be distributed to form multiple hardware or software units. Therefore, such integrated or distributed embodiments are included within the scope of the embodiments of this specification, even if not otherwise mentioned.

[0037] FIG. 1 is a schematic diagram illustrating the operational structure of a business assistance system using an AI agent according to one embodiment of the present invention, and FIG. 2 is a diagram illustrating the overall configuration of a business assistance system using an AI agent according to one embodiment of the present invention.

[0038] Referring to FIGS. 1 and 2, a work assistance system (100) using an AI agent according to one embodiment of the present invention may include a template input unit (110), a work content acquisition unit (120), an analysis information generation unit (130), and a result processing unit (140). In one embodiment, the work assistance system (100) using an AI agent may further include a user interface module (150).

[0039] The template input unit (110) receives template information that defines the format of the business result to be generated by the user, verifies and stores the input information, and can provide it so that the analysis information generation unit (130) and the result processing unit (140) can refer to it. In addition, the template information may include document type, file format identifier, page section definition, layout of body and table, field dictionary, data type and required status of each field, formatting rules and value constraints, output order, header and footer, export options, etc.

[0040] More specifically, the template input unit (110) may receive template information in a structured format such as JSON or a document file of a word processor application. The template information includes one or more fields, and each field may have metadata such as a field identifier, display name, data type, length limit, allowed value range, format string, required status, default value, interdependency condition, and whether repetition is allowed. In one embodiment, when a table-based template is specified, column definitions, header row status, cell merging rules, number format, and unit notation rules may be registered together.

[0041] In one embodiment, the template input unit (110) can perform a schema consistency check. Additionally, missing required items, data type mismatch, violation of length and pattern rules, conflicts in interdependency conditions, duplicate field identifiers, and whether section boundaries overlap can be checked in stages. At this time, if violation items exist, the location and cause of the violation may be displayed to the user and a correction input may be required, and the template may be switched to an active state only when all violations are resolved.

[0042] In one embodiment, the template input unit (110) may provide template management metadata for the template information. Additionally, a template identifier, revision number, author, creation time and modification time, usage status, and a list of supported output formats may be stored together. At this time, revision increment rules may be applied so that the change history can be preserved in a structured form.

[0043] Additionally, the template input unit (110) can calculate template feature values ​​and provide them to the analysis information generation unit (130). The feature values ​​are intended to train a large-scale language model to learn an output format corresponding to the template, and can be used to construct a prompt to be input into the large-scale language model along with the user's work content, or can be input and trained into the large-scale language model as a preliminary prompt before inputting the user's work content into the large-scale language model.

[0044] In one embodiment, the feature values ​​may consist of the order of output sections, hierarchical relationships between fields, formatting rules by field type, unit and symbol rules, arrangement of body text and tables, fixed syntax for headers and footers, hierarchical display rules for titles and subtitles, and separators and header keywords. In this case, the feature values ​​may be referenced during the output format setting step of a large-scale language model.

[0045] In addition, in one embodiment, the template input unit (110) may receive user input to select one of a plurality of templates before generating analysis information. Additionally, the selection action may include list viewing, search, filter, and preview functions, and the selected template may be bound to the current work session. In one embodiment, recently used items and favorite indicators are stored so that the re-selection process can be shortened during repetitive work.

[0046] For example, when registering a new template, the template input section (110) may support file upload or editing screen-based definition. In this case, the editing screen-based definition may be a method of instructing a specific application running on a user's computing device, such as a word processor application, to be mapped to a specific template. Additionally, when a registered template is modified, the revision may be increased and the previous revision may be converted to an archived state, and a retention policy may be applied in which a deletion request converts it to a disabled state. In this case, deletion may be restricted if there is an ongoing work session.

[0047] Additionally, the template input unit (110) can pre-calculate and provide a field map required by the result processing unit (140). Furthermore, the field map may include a correspondence between a field identifier and a physical location within the output document, a table structure index, a paragraph formatting code, and a value filling rule. At this time, the external module can fill the analysis information in sections by referring to the field map.

[0048] In one embodiment, internationalization and regional settings may be applied to the template information. Additionally, regional codes and format strings may be provided so that date, number, and currency formats match the template rules, and if the file format is specified as HWP or DOCX, format identifiers and version information may be stored in association.

[0049] At this time, an exception handling procedure may be included. Additionally, if unsupported file format identifiers or interdependency rule conflicts are detected, the template may be marked as inactive and saving may be withheld. According to one embodiment, if compatibility is degraded due to changes in field identifiers during the revision increment process, a transformation map for the previous revision may be generated so that the mapping with the old version input can be maintained during the analysis phase.

[0050] The work content acquisition unit (120) can perform the role of acquiring screen images and / or text of a computing device containing the user's work content. Additionally, the work content acquisition unit (120) can structure the acquired screen images and / or text together with format, metadata, and integrity information so that they can be utilized for subsequent processing, and provide them to an upper module.

[0051] More specifically, the work content acquisition unit (120) can detect a change in the state of an application activated on a computing device and automatically acquire screen images and / or text using an active switching event as a trigger. Additionally, the work content acquisition unit (120) can identify the active window immediately preceding the point in time when the AI ​​agent is activated by the user and automatically acquire screen images and / or text from the final screen of the window.

[0052] In one embodiment, the work content acquisition unit (120) may call the screen capture interface of the operating system to acquire a screen image and may be configured to selectively capture one of the entire screen, the active window, or a designated area. Additionally, the work content acquisition unit (120) may specify the resolution, color space, compression method, and file format to save the capture result in a standard image format or transmit it to an upper module in the form of a memory buffer.

[0053] Additionally, the work content acquisition unit (120) may prioritize legitimate text access paths, such as accessibility interfaces, clipboard reading, and application provision APIs, to directly acquire text. Additionally, if direct access to text is not possible, the work content acquisition unit (120) may supplementarily acquire text by applying an optical character recognition algorithm to a screen image.

[0054] Additionally, the work content acquisition unit (120) may create a content object according to the acquisition path and assign metadata to the object, such as a capture type, application identifier, window identifier, capture time, time zone, hash value, data size, and format identifier. In one embodiment, the work content acquisition unit (120) may perform a duplicate determination based on content hash and time criteria to prevent repeated acquisition of the same screen.

[0055] Additionally, the work content acquisition unit (120) can remove or de-identify personal identification information or unnecessary areas from the capture result by applying a predefined masking rule. Additionally, the work content acquisition unit (120) can exclude a specific coordinate range from the capture target by referring to an exception area specified by the user.

[0056] In one embodiment, the work content acquisition unit (120) may perform preprocessing on the screen image to ensure acquisition quality. Additionally, the work content acquisition unit (120) may apply tilt correction, noise removal, contrast correction, and resolution resampling to provide signal quality suitable for subsequent recognition and analysis steps.

[0057] Additionally, the work content acquisition unit (120) can perform normalization procedures such as character encoding, line break rules, space normalization, and special character matching on the text acquisition result. In one embodiment, the work content acquisition unit (120) attempts cell decomposition based on delimiters or coordinates for table-shaped text and can store the table structure separately as metadata.

[0058] Additionally, in one embodiment, the work content acquisition unit (120) can policy the trigger conditions of the acquisition procedure. Additionally, the work content acquisition unit (120) can be set to initiate the acquisition procedure only when one or more conditions are satisfied, such as activation of a specific application, acquisition of window focus, input of a shortcut key, or elapsed time.

[0059] In one embodiment, the work content acquisition unit (120) can manage acquisition failures and exception situations. Additionally, the work content acquisition unit (120) can record reasons such as insufficient authority, screen protection status, remote session restriction, and accessibility interface disabling as failure codes, and after sequentially attempting possible alternative paths, log the failure history.

[0060] Additionally, the work content acquisition unit (120) can package the acquired screen image and / or text into an input format required by the analysis information generation unit (130) and transmit it. In one embodiment, the work content acquisition unit (120) can generate a single data structure including a content body, metadata, and an integrity hash, and publish it to an upper module via a synchronous or asynchronous message queue.

[0061] In addition, in one embodiment, the work content acquisition unit (120) may reflect user settings and security policies. In addition, the work content acquisition unit (120) may receive a list of applications allowed to capture, sensitive information masking rules, storage location, retention period, and encryption status as setting values ​​and apply them throughout the acquisition process.

[0062] The analysis information generation unit (130) can generate analysis information corresponding to the template structure based on the template feature value provided from the template input unit (110) and the screen image and / or text transmitted from the work content acquisition unit (120). That is, the analysis information can be configured as a prompt consisting of the template specific value and the user's work screen image and / or text sequentially or simultaneously. Additionally, the analysis information generation unit (130) can apply to a large-scale language model by configuring the order of the output target fields and sections, key names, value formats, units, date and number notation rules as output control specifications.

[0063] More specifically, the analysis information generation unit (130) can combine three input packages during the input stage. Additionally, the analysis information generation unit (130) can merge the template feature value package, the work content package, and the session metadata package, and sort the merged results according to an internal standard schema. In one embodiment, the analysis information generation unit (130) can extract a section ID, field identifier, required status, format string, allowed value, unit rule, and sequence relationship between sections from the template feature value, and extract body text, table candidates, keyword occurrence locations, and citation ranges from the work content.

[0064] Meanwhile, the analysis information generation unit (130) can perform a procedure for setting the output format for a large-scale language model. Additionally, the analysis information generation unit (130) can assemble an output control block consisting of system directives, schema declarations, fixed syntax, prohibited syntax, a list of required fields, and example instances, and place the assembled block at the beginning of the model input. In one embodiment, the analysis information generation unit (130) can specify a JSON schema and key order, instruct to maintain an empty key when required fields are missing, and specify rules such as the date field being in year-month-day format and the amount field being returned only as a number.

[0065] Next, the analysis information generation unit (130) can control model call parameters. Additionally, the analysis information generation unit (130) can suppress unexpected insertion of descriptive sentences or schema deviations by setting sampling intensity, maximum number of tokens, response interruption tokens, and prohibition tokens. In one embodiment, the analysis information generation unit (130) can enforce a separator token at the end of a section and specify whether line breaks within field values ​​are allowed for each field type.

[0066] In addition, the analysis information generation unit (130) can perform format verification and content verification on the first generation result. Furthermore, the analysis information generation unit (130) can sequentially perform JSON parser verification, schema conformance verification, regular expression-based pattern verification, unit conversion possibility check, and date and number notation rule check. In one embodiment, the analysis information generation unit (130) can configure partial regeneration instructions targeting only the fields where discrepancies are found, and request the model to correct and output only the corresponding fields in the same context.

[0067] Additionally, the analysis information generation unit (130) can reflect sensitive information and security policies. Additionally, the analysis information generation unit (130) can replace field values ​​corresponding to masking rules with separate tokens and record only hashes in the logs to maintain the reproduction path while restricting the exposure of the original text. In one embodiment, the analysis information generation unit (130) can apply a pre-filter and a post-filter to both the model input and output based on personal identification information detection patterns and a list of prohibited words.

[0068] Additionally, the analysis information generation unit (130) can process internationalization settings and region settings. Additionally, the analysis information generation unit (130) can convert date formats, number separators, and currency notations according to the locale code, and insert fixed phrases for titles, headers, and footers in the language specified in the template feature value. In one embodiment, if multilingual output is required for the same template, the analysis information generation unit (130) can generate language-specific schema derivatives to maintain the same partition structure.

[0069] The result processing unit (140) receives as input the template feature values ​​provided by the template input unit (110) and the analysis information transmitted from the analysis information generation unit (130), and can configure the business result data according to the sections, fields, and formatting rules defined in the template feature values. In addition, the result processing unit (140) can provide the configured business result data by arranging it into an internal standard schema that can be directly referenced by the user interface module and the export module.

[0070] Meanwhile, the result processing unit (140) can reflect a sensitive information processing policy. Additionally, the result processing unit (140) can apply masking rules to specific fields and insert non-identification tokens for values ​​that need to be excluded from export. At this time, the result processing unit (140) can record only hash values ​​and location references in an internal log in preparation for the need to restore the original text.

[0071] Additionally, the result processing unit (140) can adjust the output format by taking into account internationalization and regional settings. In one embodiment, the result processing unit (140) can apply a date separator, a decimal point and a thousands separator, and a currency notation according to the locale code, and insert a language-specific fixed syntax defined in the template.

[0072] Next, the result processing unit (140) can prepare for export to a specified file format through an output adapter. Additionally, if a file format (e.g., HWP or DOCX, etc.) to be used in a specific application is specified, the result processing unit (140) can map paragraph styles, table styles, fonts, margins, headers, and footers to file format objects, and convert character encoding and line break rules according to the format specifications.

[0073] In one embodiment, the result processing unit (140) may provide a preview rendering for linkage with a user interface module. Additionally, the result processing unit (140) may generate preview data including field-unit highlight information and provide difference information so that only the changed area can be partially updated when modifications are reflected.

[0074] Additionally, in one embodiment, the result processing unit (140) may record the processing history. Additionally, the result processing unit (140) may store the template revision used, field map version, batch rule version, export format version, creation time, and processing step summary as session metadata.

[0075] The user interface module (150) can display work result data provided by the result processing unit (140) on the screen and provide input components so that the user can modify values ​​by field and then instruct the saving to a specified file format. Additionally, the user interface module (150) can visually distinguish between an editable area and a fixed syntax area by referencing template feature values ​​transmitted from the template input unit (110), and apply field-specific formatting rules and constraints to the editing operation.

[0076] More specifically, the user interface module (150) can organize the screen into areas to arrange a document preview area, a field attribute editing area, and an export control area that reflect the template structure. Additionally, the user interface module (150) can render the body, table, header, footer, and title hierarchy in the document preview area in the order of the template definition, and combine a focusable input widget with an editable field.

[0077] In one embodiment, the user interface module (150) can automatically map widgets corresponding to field types, such as text fields, number inputs, date pickers, selection lists, and table cell editors. Additionally, the user interface module (150) can perform validation at the time of input by reflecting whether it is required, a range of allowed values, a format string, and a length limit as widget attributes.

[0078] Meanwhile, the user interface module (150) can detect an editing event, collect the changes in increments, and send them to the result processing unit (140) to request a preview re-rendering.

[0079] Additionally, the user interface module (150) can apply internationalization and regional settings to convert date, number, currency notation and fixed syntax language during the screen rendering stage. In one embodiment, the user interface module (150) can maintain the section structure of the template derivative when switching languages ​​and update only the values ​​according to localization rules.

[0080] Additionally, the user interface module (150) can reflect security and privacy protection policies. Additionally, the user interface module (150) can display the masking target field as a separate token when rendering and block the display of the original text in sessions without restoration rights. At this time, the user interface module (150) can insert a security watermark or session identifier as an overlay when copying the screen and downloading.

[0081] FIG. 3 is a schematic diagram illustrating the hardware configuration of a work assistance system (100) using an AI agent according to one embodiment of the present invention.

[0082] Referring to FIG. 3, a work assistance system (100) using an AI agent according to embodiments is implemented in the form of a computing device including hardware (200) and may include memory (210), a processor (220), a communication module (230), and an input / output unit (240).

[0083] The memory (210) is a non-transient computer-readable recording medium and may include a permanent mass storage device such as RAM (random access memory), ROM (read only memory), disk drive, SSD (solid state drive), flash memory, etc. Here, the permanent mass storage device such as ROM, SSD, flash memory, disk drive, etc. may be included in the device or server described above as a separate permanent storage device distinct from the memory (210).

[0084] Additionally, the memory (210) may store an operating system and at least one program code (e.g., code for a security module or an application installed to provide a specific service). These software components may be loaded from a computer-readable recording medium separate from the memory (210). This separate computer-readable recording medium may include a computer-readable recording medium such as a floppy drive, disk, tape, DVD / CD-ROM drive, or memory card.

[0085] In another embodiment, software components may be loaded into memory (210) via a communication module (230) rather than a computer-readable recording medium. For example, at least one program may be loaded into memory (210) based on a computer program installed by files provided over a network by developers or a file distribution system (e.g., an application store service server) that distributes installation files for applications.

[0086] The processor (220) may be configured to process instructions of a computer program by performing basic arithmetic, logic, and input / output operations. Instructions may be provided to the processor (220) by memory (210) or a communication module (230). For example, the processor (220) may be configured to execute instructions received according to program code stored in a recording device such as memory (210).

[0087] The communication module (230) can provide a function for communicating with a user terminal, etc., through a network. Additionally, the communication module (230) can provide a function for communicating with one or more other devices through a wired or wireless network. That is, the communication module (230) is a part that realizes each function module described above with reference to FIG. 2 by controlling its function by a processor (220) that references memory (210).

[0088] The input / output unit (240) may be a means for interfacing with an external input / output device (not shown). For example, the external input device may include devices such as a keyboard, mouse, microphone, camera, etc., and the external output device may include devices such as a display, speaker, haptic feedback device, etc. As another example, the input / output unit (240) may be a means for interfacing with a device in which the functions for input and output are integrated into one, such as a touchscreen.

[0089] In addition, in other embodiments, the work assistance system (100) using an AI agent may include more hardware components than those shown in FIG. 3 depending on the nature of the device to which it is applied. For example, it may be implemented to include at least some of the input / output devices described above, or it may include additional components such as a transceiver, a GPS (Global Positioning System) module, a camera, various sensors, a DB, etc. As a more specific example, it may be implemented to include various additional components such as an accelerometer or gyroscope sensor, a camera module, various physical buttons, buttons using a touch panel, input / output ports, and a vibrator for vibration.

[0090] However, the components and forms of the computing device described in this specification are merely exemplary, and the configuration of the computing device in which the work assistance system (100) using an AI agent is implemented may differ from that described in this specification depending on the adoption of other known technologies or future advancements in information and communication technology.

[0091] Next, we will look at the process of receiving template information that defines the format of the work result through the template input section (110).

[0092] FIG. 4 is a flowchart illustrating the process of receiving template information that defines the format of a business result through a template input unit (110) according to one embodiment of the present invention.

[0093] Referring to FIG. 4, first, in step S401, the template input unit (110) can initialize the template input screen and load the current work session identifier and previous usage history. Additionally, the template input unit (110) can load a list of recently used templates and favorite information and display them in the selection area.

[0094] Next, in step S402, the template input unit (110) can receive the user's selection input and branch whether to select an existing template or register a new template. In one embodiment, the template input unit (110) may display a file format selection item together to require the user to first specify one of the predefined file formats (e.g., HWP or DOCX).

[0095] Next, in step S403, the template input unit (110) can receive a selected template file via upload or receive field information defined via screen-based editing. Additionally, the template input unit (110) can parse the received structure into an internal standard schema and extract field identifiers, data types, required status, format strings, and unit rules.

[0096] Next, in step S404, the template input unit (110) can perform schema normalization to convert the output order of the section unit, table area definitions, header and footer syntax, and page section boundaries into an internal representation. Additionally, the template input unit (110) can record the dependency relationships between fields and default value rules in a mapping table.

[0097] Next, in step S405, the template input unit (110) can perform a consistency check to verify whether required items are missing, data types are mismatched, length and pattern rules are violated, field identifiers are duplicated, or section boundaries overlap. In one embodiment, the template input unit (110) can list the inspection results by item and generate the location of the violation and the reason for the violation together.

[0098] Next, in step S406, the template input unit (110) may request correction input if violation items exist and perform a re-examination on the modified schema. Additionally, the template input unit (110) may keep the active switching in a suspended state until the iterative correction is completed.

[0099] Next, in step S407, the template input unit (110) may assign and store management metadata, such as a template identifier, revision number, author, creation time, and modification time, to the template that has passed the consistency check. In one embodiment, the template input unit (110) may apply an auto-incrementing revision policy to prevent identical name conflicts.

[0100] Next, in step S408, the template input unit (110) can calculate template feature values ​​to be used in the output format setting and result processing steps of the large-scale language model. Additionally, the template input unit (110) can package section order, field hierarchy, formatting rules, unit and symbol rules, table column definitions, and fixed syntax as feature values.

[0101] Next, in step S409, the template input unit (110) can bind the selected template to the current work session and set the result export format as a session attribute. In one embodiment, the template input unit (110) can record the binding information at the end of the session so that it can be reused in a subsequent session.

[0102] Finally, in step S410, the template input unit (110) can display feature values ​​and a field map summary along with a preview for verification and receive user approval input. Additionally, the template input unit (110) can notify the session of the template activation status upon approval completion so that it can be referenced in a subsequent analysis step.

[0103] Next, we will examine the process by which the work content acquisition unit (120) acquires the screen image and / or text of the computing device.

[0104] FIG. 5 is a flowchart illustrating the process of a work content acquisition unit (120) according to an embodiment of the present invention acquiring a screen image and / or text of a computing device.

[0105] Referring to FIG. 5, the work content acquisition unit (120) can initialize the acquisition policy and register a session identifier, an application watchlist, a window focus event, and a shortcut key trigger. Additionally, the work content acquisition unit (120) can load recent acquisition history and duplicate determination criteria.

[0106] Next, in step S502, the work content acquisition unit (120) can detect the activation of a preset application or the switching of an AI agent's activation, and determine the window identifier and application identifier of the immediately active window. Additionally, the work content acquisition unit (120) can select a mode to determine whether the acquisition target is the entire screen, the active window, or a designated area.

[0107] Next, in step S503, the work content acquisition unit (120) determines whether direct text acquisition is possible, and if one or more of the accessibility interface, application API, and clipboard path are available, the text path may be selected first. Additionally, if the text path is not available, the work content acquisition unit (120) may branch to an image path.

[0108] Next, in step S504, the work content acquisition unit (120) can call the operating system capture interface to acquire a screen image and set the resolution, color space, compression method, and file format. Additionally, the work content acquisition unit (120) can secure the capture result into a memory buffer and calculate an integrity hash.

[0109] Next, in step S505, the work content acquisition unit (120) can perform preprocessing such as tilt correction, noise removal, brightness correction, and resampling on the acquired screen image. Additionally, the work content acquisition unit (120) can de-identify personal identification areas or non-acquired areas according to a predefined masking rule.

[0110] Next, in step S506, the work content acquisition unit (120) can extract text by performing OCR processing if the text is not directly obtained. Additionally, the work content acquisition unit (120) can set a language model, dictionary, and paragraph splitting rules, and calculate character coordinates and reliability together.

[0111] Next, in step S507, the work content acquisition unit (120) can perform character encoding normalization, line break alignment, space cleanup, and special character unification on the string collected from the text direct acquisition path. Additionally, the work content acquisition unit (120) can attempt delimiter-based or coordinate-based cell decomposition for the table candidate text.

[0112] Next, in step S508, the work content acquisition unit (120) can create an acquired content object and assign metadata such as an acquisition mode, application identifier, window identifier, capture time and time zone, data size, format identifier, and integrity hash. Additionally, the work content acquisition unit (120) can record whether masking is applied and the rule version.

[0113] Next, in step S509, the work content acquisition unit (120) can determine whether there is a duplicate by comparing the content hash and time interval with the previous acquired version. Additionally, if the work content acquisition unit (120) determines that there is a duplicate, it may omit reacquisition or extract only the changed section.

[0114] Next, in step S510, the work content acquisition unit (120) checks the quality criteria and, if the resolution threshold, blur index, and text coverage ratio do not meet the criteria, it may attempt reacquisition or an alternative path. Additionally, the work content acquisition unit (120) may count the reason for failure and the number of retries.

[0115] Next, in step S511, the work content acquisition unit (120) can package the body, metadata, and integrity information according to the input schema required by the analysis information generation unit (130). Additionally, the work content acquisition unit (120) can deliver the package via a synchronous call or message queue and record an acknowledgment of receipt.

[0116] Finally, in step S512, the work content acquisition unit (120) can perform exception handling and security procedures. Additionally, the work content acquisition unit (120) can record failure codes for insufficient authority, screen protection status, remote session restriction, and capture blocking policy, clear temporary buffers, and apply storage retention periods and encryption policies.

[0117] Next, we will examine the process in which the work content acquisition unit (120) automatically acquires screen images and / or text in response to the activation of a preset application.

[0118] FIG. 6 is a flowchart illustrating the process of a work content acquisition unit (120) according to an embodiment of the present invention automatically acquiring a screen image and / or text in response to a preset application activation.

[0119] Referring to FIG. 6, first, in step S601, the work content acquisition unit (120) can initialize an automatic acquisition policy. Additionally, the work content acquisition unit (120) can load the application allow list, window focus monitoring cycle, minimum dwell time threshold, number of retries and interval, capture mode default value (one of full screen, active window, or designated area), image format default value and resolution, and text priority status into the settings and generate a session identifier. In one embodiment, the work content acquisition unit (120) can verify screen capture permission, accessibility interface usage permission, and clipboard access permission by performing a permission check procedure beforehand.

[0120] Next, in step S602, the task content acquisition unit (120) can register an operating system event hook. Additionally, the task content acquisition unit (120) can be configured to receive application activation events, focus switching events, window creation and destruction events, and agent activation signals, and can write the handle, process identifier, class name, and timestamp of the immediately active window to a circular buffer.

[0121] Next, in step S603, the work content acquisition unit (120) can determine the information of the immediately preceding active window as soon as it detects a transition to the agent window. Additionally, the work content acquisition unit (120) can filter out invalid triggers to initiate automatic acquisition only when the transition time and the dwell time of the immediately preceding window are calculated and the minimum dwell time threshold is satisfied. In one embodiment, the work content acquisition unit (120) can perform capture after window synthesis stabilization by setting a certain time delay value (e.g., 100 to 300 milliseconds) after the transition.

[0122] Next, in step S604, the task content acquisition unit (120) can determine whether the immediately active window is included in the allow list. Additionally, the task content acquisition unit (120) evaluates whether to allow based on the process name, signature issuer, window class name, and window property flag (whether to display secure content), and if not allowed, stops automatic acquisition and logs with a reason code. In one embodiment, the task content acquisition unit (120) can provide an exception allowance flow by displaying a confirmation dialog box if the user settings allow it.

[0123] Next, in step S605, the work content acquisition unit (120) can determine whether direct access to text is possible. Additionally, the work content acquisition unit (120) checks whether an accessibility interface is supported, whether an application API is provided, clipboard ownership, and data format, and can exclude the text path if a DRM or capture blocking flag is set. In one embodiment, the work content acquisition unit (120) can apply a text priority policy if the text path is possible, and branch to an image path if it is not possible.

[0124] Next, in step S606, the work content acquisition unit (120) can perform direct text acquisition. Additionally, the work content acquisition unit (120) can collect strings using available paths among reading text ranges of an accessibility interface, querying document models of an application API, and extracting plain text from the clipboard, and can apply character encoding unification, line break rule alignment, whitespace normalization, and control character removal. In one embodiment, if a hierarchical structure is provided, the work content acquisition unit (120) can collect section IDs, paragraph indices, and correspondence information with table cell coordinates together.

[0125] Next, in step S607, the work content acquisition unit (120) can perform an image acquisition path. Additionally, the work content acquisition unit (120) can call an operating system capture API to set the target area to an active window or specified coordinates, and specify the resolution, scale factor, color space, compression method, and file format. In one embodiment, the work content acquisition unit (120) can apply options such as excluding window decorations and shadows, excluding transparent window composites, and multi-monitor coordinate correction.

[0126] Next, in step S608, the work content acquisition unit (120) can acquire supplementary text through OCR processing. Additionally, the work content acquisition unit (120) can set a language model, dictionary, binarization threshold, slope correction, and extinction line removal parameters, and calculate character coordinates, line range, and confidence score together for the recognition result. In one embodiment, the work content acquisition unit (120) can detect table structure candidates to estimate row and column boundaries and divide the text using coordinate-based cell clues.

[0127] Next, in step S609, the work content acquisition unit (120) can perform masking and exception coordinate processing. Additionally, the work content acquisition unit (120) can refer to the personal identification information pattern and the user-specified non-acquisition area to blur or mosaic the area, and record the masking map and rule version identifier in the metadata. In one embodiment, the work content acquisition unit (120) can replace sensitive tokens with replacement strings in the case of text paths.

[0128] Next, in step S610, the work content acquisition unit (120) can create an acquired content object. Additionally, the work content acquisition unit (120) can assign metadata to the object, such as an acquisition path type (text or image), an application identifier, a window identifier, an acquisition time and time zone, a data size, a format identifier, a hash value (e.g., SHA-256, etc.), and quality indicators (resolution, blur index, text coverage). In one embodiment, the work content acquisition unit (120) can apply encryption upon storage and specify a retention period and an access level.

[0129] Next, in step S611, the work content acquisition unit (120) can perform a duplicate determination. Additionally, the work content acquisition unit (120) can compare with the previous acquired version using a similarity index of a normalized string in the case of text and a perceptual hash in the case of an image, and if it is determined that the difference is less than a threshold, it can omit reacquisition or extract only the changed area. In one embodiment, the work content acquisition unit (120) can log the reason for omitting duplicates and the comparison indicators.

[0130] Finally, in step S612, the work content acquisition unit (120) may perform data packaging and transmission for linkage with the analysis information generation step. Additionally, the work content acquisition unit (120) may serialize into an internal schema including a body payload, metadata, and an integrity hash, transmit via a synchronous call or message queue, and record an acknowledgment of receipt. In one embodiment, the work content acquisition unit (120) may log the failure code, the number of retries, and the elapsed time, and clean up the session by securely clearing the temporary buffer and the keyboard input capture buffer.

[0131] Next, we will examine the process of generating analysis information by inputting an image and / or text of a work screen into a large-scale language model in which the output format is set in advance by the analysis information generation unit (130).

[0132] FIG. 7 is a flowchart illustrating the process of generating analysis information by inputting an image and / or text of a work screen into a large-scale language model in which the output format is set in advance, by an analysis information generation unit (130) according to one embodiment of the present invention.

[0133] Referring to FIG. 7, first, in step S701, the analysis information generation unit (130) may receive template feature values ​​from the template input unit (110) and receive screen images and / or text and their metadata from the work content acquisition unit (120). Additionally, the analysis information generation unit (130) may initialize session identifiers, locales, unit rules, and date notation rules, and align input packages to an internal standard schema.

[0134] Next, in step S702, the analysis information generation unit (130) can assemble an output format control block. Additionally, the analysis information generation unit (130) can extract a partition order, field identifier, mandatory status, format string, allowed value, unit and symbol rule, and fixed syntax from the template feature value to construct a control block including a schema declaration and prohibited syntax, a list of required fields, and example instances.

[0135] Next, in step S703, the analysis information generation unit (130) can determine a splitting policy by evaluating the input length and structure. Additionally, the analysis information generation unit (130) can split the text based on the section boundary when the token limit is exceeded, assign an overlapping section, and map a section identifier to each split. In one embodiment, the analysis information generation unit (130) can ensure table restoration accuracy by maintaining coordinate-based cell clues in the image-derived text.

[0136] Next, in step S704, the analysis information generation unit (130) can configure the model input context. Additionally, the analysis information generation unit (130) can place a control block at the beginning, combine the divided work content fragments and session metadata in sequence, and specify a separator and a break token to set the output for each section to be separated.

[0137] Next, in step S705, the analysis information generation unit (130) can set model call parameters. Additionally, the analysis information generation unit (130) can specify the sampling intensity, maximum number of tokens, stop tokens, and prohibit tokens, and select whether to force JSON format options and deterministic regeneration mode.

[0138] Next, in step S706, the analysis information generation unit (130) can perform primary generation for each split input. Additionally, the analysis information generation unit (130) can preserve the response body, model identifier, prompt version, and parameter summary together.

[0139] Next, in step S707, the analysis information generation unit (130) can perform format verification. Additionally, the analysis information generation unit (130) can perform checks for JSON parsing, schema suitability, existence of required keys, string patterns, date formats, and number conversion possibilities in stages.

[0140] Next, in step S708, the analysis information generation unit (130) can perform content verification. Additionally, the analysis information generation unit (130) can check for dependency conditions between fields, value ranges, unit conversion rules, and duplicate key conflicts, and list the violation items.

[0141] Next, in step S709, the analysis information generation unit (130) can perform a partial regeneration loop. Additionally, the analysis information generation unit (130) can configure a correction instruction limited to the violation field to request that only the corresponding field be re-output in the same context, and repeat this until the maximum number of retries is reached.

[0142] Next, in step S710, the analysis information generation unit (130) can perform partition merging and sorting. Additionally, the analysis information generation unit (130) can merge the results for each partitioned piece using the partition ID and field identifier as keys, and apply a merging policy to duplicate fields based on the occurrence location, confidence score, and latest modification time.

[0143] Next, in step S711, the analysis information generation unit (130) can generate reliability and evidence tracking information. Additionally, the analysis information generation unit (130) can calculate field-specific reliability based on schema suitability, keyword matching, section coverage, and conflict detection, and attach the original text citation range and partition fragment identifier based on the grounds.

[0144] Next, in step S712, the analysis information generation unit (130) can apply security policies and sensitive information processing. Additionally, the analysis information generation unit (130) can limit the exposure of the original text by replacing the masking target value with a token and recording only the hash in the log.

[0145] Finally, in step S713, the analysis information generation unit (130) can package the final analysis information into an internal standard schema and transmit it to the result processing unit (140), and store the session summary, error list, and regeneration trace. Additionally, the analysis information generation unit (130) can clear the temporary buffer and terminate the processing state after receiving a delivery confirmation.

[0146] Next, we will examine the process in which the template input unit (110) receives a template selected by the user from among a plurality of template information before generating analysis information.

[0147] FIG. 8 is a flowchart illustrating the process of a template input unit (110) according to an embodiment of the present invention receiving a template selected by a user among a plurality of template information before generating analysis information.

[0148] Referring to FIG. 8, first, in step S801, the template input unit (110) can initialize the template selection screen and load summary information including the template list, classification, file format, last modified time, revision, and usage status. Additionally, the template input unit (110) can configure the list by applying a search term, classification filter, format filter, and sorting criteria.

[0149] Next, in step S802, the template input section (110) can generate a preview of the list item and display a summary of the document structure, the number of required fields, the presence of a table area, and an exportable format. Additionally, the template input section (110) can expand a field tree, formatting rules, and version history in a detailed properties panel when an item is selected.

[0150] Next, in step S803, the template input unit (110) receives the user's template selection input and can check the access rights and usage status of the selected item. Additionally, if the template input unit (110) detects a disabled state or insufficient permission, it can block the selection and display a reason code.

[0151] Next, in step S804, the template input unit (110) can confirm the revision of the selected template and verify compatibility with the version required by the project or session. Additionally, if there is a revision conflict, the template input unit (110) can suggest a recommended revision or apply a transformation map to resolve differences in field identifiers.

[0152] Next, in step S805, the template input unit (110) can check for compatibility with the session settings and verify whether there is a match with the locale, unit rules, date notation rules, and export target format. Additionally, if there are discrepancies, the template input unit (110) can adjust session-side parameters or template-side options through user confirmation.

[0153] Next, in step S806, the template input unit (110) can bind the selected template to the current work session and record the template identifier, revision, and output format in the session attributes. Additionally, the template input unit (110) can set a session lock indicator on the selected item to prevent simultaneous editing conflicts.

[0154] Next, in step S807, the template input unit (110) can calculate the template feature values ​​and field map summary and cache them in a form that can be referenced in the analysis step and result processing step. Additionally, the template input unit (110) can separately extract the list of required fields and value constraint rules and pass them to the subsequent validation flow.

[0155] Finally, in step S808, the template input unit (110) can display the binding result and feature value summary as a preview and receive user approval input. Additionally, after approval, the template input unit (110) can save the selection history and time information as an audit log and send a selection completion signal to the analysis information generation unit (130) to initiate subsequent processing.

[0156] Next, we will look at the process of displaying the work result data generated through the user interface module (150) on the screen and downloading it in a specified file format after the user modifies it.

[0157] FIG. 9 is a diagram illustrating the process of displaying business result data generated through a user interface module (150) according to an embodiment of the present invention, and the user downloading it in a specified file format after modification.

[0158] Referring to FIG. 9, first, in step S901, the user interface module (150) can load session information and check the current template identifier, revision, and output format settings. Additionally, the user interface module (150) can receive work result data and field map summary transmitted from the result processing unit (140) and switch to an initial rendering ready state.

[0159] Next, in step S902, the user interface module (150) can initialize the document preview area and render the body, table, header, footer, and title hierarchy according to the template definition order. Additionally, the user interface module (150) can assign an input cursor and border indicator to editable fields and apply an uneditable indicator to fixed phrases.

[0160] Next, in step S903, the user interface module (150) can map a widget corresponding to a field type. Additionally, the user interface module (150) can place text fields, number inputs, date pickers, selection lists, table cell editors, etc., and apply requirements, length limits, format strings, and allowed value ranges as widget constraints.

[0161] Next, in step S904, the user interface module (150) can detect user edit events and collect change values ​​in differential units. Additionally, the user interface module (150) can prevent excessive recalculation by applying debouncing and throttling policies, and request a partial preview update by sending only the changed fields to the result processing unit (140).

[0162] Next, in step S905, the user interface module (150) can perform input validation. Additionally, if the user interface module (150) detects a format mismatch, missing required fields, or unit conversion failure, it can display a warning mark in the corresponding area and provide correction instructions via a tooltip. In one embodiment, the user interface module (150) can disable export control when validation is not passed.

[0163] Next, in step S906, the user interface module (150) can support editing the table structure. Additionally, the user interface module (150) can provide functions for adding or deleting rows, adjusting column order, and unmerging cells, but can restrict editing of columns and header rows that are fixed in the template.

[0164] Next, in step S907, the user interface module (150) can reflect the preview update response. Additionally, the user interface module (150) can re-render only the changed section by applying the difference information provided by the result processing unit (140), and maintain the scroll position and focus to ensure editing continuity.

[0165] Next, in step S908, the user interface module (150) may display an export dialog box for saving to a specified file format. Additionally, the user interface module (150) may confirm the file format, document title, version name, and save location, and after sending the export request, display the processing progress and completion status.

[0166] Finally, in step S909, the user interface module (150) can provide the save result to the user and record the session history. Additionally, the user interface module (150) can display a download link or save path and leave the change field, previous value and change value, save time, and file format information as an audit log.

[0167] Next, we will examine the UI configuration for template selection and automatic acquisition of work content according to one embodiment of the present invention.

[0168] FIG. 10 is a drawing illustrating an example of a template selection and automatic acquisition of work content UI according to an embodiment of the present invention.

[0169] Referring to FIG. 10, in one embodiment, a template list may be displayed in the left area of ​​the screen. In this case, the template list may be configured to list a plurality of previously registered templates by item, and the user may check detailed information by clicking or selecting each item. In one embodiment, each template item may be configured to include a text label and a selectable state, and a highlighting process that is visually distinguishable when selected may be applied.

[0170] Additionally, detailed information of the selected template may be displayed on the right side of the template list. In this case, the detailed information may include attributes such as the template's file format, the number of fields included, the last modified date and time, the author, formatting rules, and exportable formats. In one embodiment, buttons capable of performing template registration, modification, and deletion functions may be placed at the bottom of the detailed information area, and each button may be configured to trigger a call to the corresponding function.

[0171] Additionally, an automatic task content acquisition status window may be displayed in the upper right corner of the screen. A list of monitored applications may be listed in the left area of ​​the automatic task content acquisition status window, and pre-set application names such as Microsoft Word, Hangul (HWP), and Google Docs may be displayed. Information on currently applied capture settings may be displayed in the right area, and items such as mode (full screen, active window, designated area), file format, resolution, whether to prioritize text, and whether to use OCR supplementation may be included.

[0172] Additionally, the history of automatic acquisition of recent work content may be displayed in a log format in chronological order at the bottom of the screen. Each log item may include the time of acquisition, the target of acquisition, the acquisition method (e.g., direct acquisition, acquisition supplemented by OCR), and information regarding success or status. In one embodiment, log items are displayed at the top in reverse chronological order and may include status messages such as the omission of duplicate acquisitions or attempts to re-acquire due to substandard quality.

[0173] Based on this UI configuration, users can quickly select a desired template, check the properties of the selected template, and simultaneously monitor the status of automatic acquisition of work content in real time.

[0174] Next, we will examine a user terminal screen configured to allow users to view and edit work result data within the same screen and download it in a specified file format, without the need for a separate copy and paste process.

[0175] FIG. 11 is a drawing illustrating a result display, editing, and download UI configuration as an example of a user terminal screen according to an embodiment of the present invention.

[0176] Referring to FIG. 11, in one embodiment, the left area represents a work screen that a user previously performed in a word processor or web-based document creation environment. In the present invention, the user can transmit the content directly to an AI agent from the current work screen without having to separately copy existing work content and paste it into an LLM (Large-scale Language Model) input window in order to utilize LLM. This reduces unnecessary screen switching and repetitive copy or paste processes.

[0177] Additionally, the right area is a modification page popup window provided by the user interface module (150) of the present invention, which displays work result data generated by the AI ​​agent. The user can immediately edit the result in this popup window, and in particular, by providing a UI similar to a Hangul (HWP) environment, the result can be easily corrected into a document format mainly required in an enterprise environment.

[0178] Additionally, a download list is displayed at the top of the pop-up window, allowing users to view previous versions or results in different formats generated by the AI ​​agent and select the desired file to download. The download supports formats frequently used in the corporate world, such as HWP and DOCX, enabling the immediate use of LLM results for work without the need for separate conversion.

[0179] In addition, control buttons such as 'Save', 'Download HWP', and 'Download DOCX' are placed at the bottom of the popup window, allowing the user to reflect the modified result in the system or immediately download it in the desired format. As such, the UI illustrated in FIG. 10 can implement a configuration that minimizes inconvenience in the process of utilizing LLM and maximizes work efficiency by providing integrated result display, real-time Hangul editing, and downloading in various formats within the web environment of the same page that was queried.

[0180] Next, we will examine the overall procedure of a task assistance method using an AI agent according to one embodiment of the present invention.

[0181] FIG. 12 is a flowchart illustrating the overall procedure of a task assistance method using an AI agent according to one embodiment of the present invention.

[0182] Referring to FIG. 12, first, in step S1201, the template input unit may receive template information that defines the format of the business result desired by the user. Additionally, the template input unit may receive items such as file format, partition structure, field list, formatting rules, value constraints, and output order in a structured format, and perform schema consistency checks and identifier duplicate checks to save them as an active template. In one embodiment, the template input unit may calculate template feature values ​​and cache them so that they can be referenced in subsequent steps.

[0183] Next, in step S1202, the task content acquisition unit may acquire a screen image and / or text of a computing device containing the user's task content. Additionally, the task content acquisition unit may detect triggers, such as an application activation switch or shortcut key input, to capture a screen image of the active window or directly extract text through an accessibility interface and an application API. In one embodiment, if direct acquisition of text is not possible, the task content acquisition unit may apply OCR processing to supplementarily acquire the text and assign metadata such as an acquisition path, window identifier, time, and hash value.

[0184] Next, in step S1203, the template input unit may receive a user's selection input for one of a plurality of templates prior to generating analysis information. Additionally, the template input unit may check the compatibility of the selected revision and bind the template identifier, revision, and output format to session attributes. In one embodiment, the template input unit may separately extract the list of required fields and the value constraint summary and transmit them to the next step.

[0185] Next, in step S1204, the analysis information generation unit may combine template feature values ​​and work content to set the output format of a large-scale language model and generate analysis information by inputting images and / or text of the work screen. Additionally, the analysis information generation unit may assemble a control block consisting of a schema declaration, fixed syntax, prohibited syntax, and a list of required keys, and perform model calling by applying section unit division and overlapping sections according to the input length. In one embodiment, the analysis information generation unit may perform JSON parsing, schema conformance, and verification of date and number notation rules on the primary generation result, and correct format deviations through partial regeneration limited to the violating fields.

[0186] Next, in step S1205, the result processing unit can convert the generated analysis information into business result data corresponding to the template information. Additionally, the result processing unit can interpret the field map to sequentially perform value normalization, unit conversion, application of paragraph and table formatting, page section arrangement, and merging of duplicate fields, and align them according to the internal standard schema required by the export adapter. In one embodiment, the result processing unit can apply image placement, table header repetition, and page break rules, and prepare mapping data to HWP or DOCX according to the format-specific style map.

[0187] Finally, in step S1206, the user interface module may display work result data on the screen and provide an edit widget to allow the user to modify it. Additionally, the user interface module may display input validation results in real time and perform a preview update that reflects only the differences of the modified content. In one embodiment, the user interface module may transmit a save request to the result processing unit and provide download control to a specified file format to save an HWP or DOCX file to the user terminal.

[0188] Through this, the present invention enables users to check, modify, and save results on the same screen without the need for separate copy, paste, or format conversion processes to utilize LLM in their work, thereby increasing work efficiency and providing the advantage of quickly delivering results in a specified format required in an enterprise environment.

[0189] The operation according to the method of the embodiments described above may be implemented at least partially as a computer program and recorded on a computer-readable recording medium. A computer-readable recording medium on which a program for implementing the operation according to the embodiments is recorded includes all types of recording devices in which data readable by a computer is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc. Additionally, the computer-readable recording medium may be distributed across networked computer systems, and computer-readable code may be stored and executed in a distributed manner. Furthermore, functional programs, codes, and code segments for implementing the present embodiment will be easily understood by a person skilled in the art to which the present embodiment belongs.

[0190] Additionally, each block or each step illustrated in the flowcharts of this specification may represent a module, segment, or part of code comprising one or more executable instructions for executing a specified logical function(s). Furthermore, in some alternative embodiments, the functions mentioned in the blocks or steps may occur out of order. For example, two blocks or steps illustrated in succession may actually be performed substantially simultaneously, or the blocks or steps may occasionally be performed in reverse order according to the corresponding function.

[0191] Although the present invention has been described in relation to some embodiments, various modifications and changes may be made without departing from the scope of the invention as understood by a person skilled in the art to which the invention pertains. Furthermore, such modifications and changes should be considered to fall within the scope of the claims appended to this specification. Explanation of the symbols

[0192] 100: Task assistance system using AI agents 110: Template Input Section 120: Work Content Acquisition Section 130: Analysis Information Generation Unit 140: Result processing section 150: User Interface Module

Claims

Claim 1 A method for assisting work using an AI agent executed by a computing device, comprising: receiving template information defining the format of a work result; a step in which an AI agent executed on the computing device acquires one or more of a screen image and text of the computing device containing the user's work content; a step of generating analysis information for a work using a feature value extracted from the template information and one or more of the screen image and text as input values ​​for a pre-trained large-scale language model; and a step of processing the analysis information into work result data corresponding to the template information, wherein the step of acquiring one or more of the screen image and text includes: a step of registering an application monitoring list; a step of detecting that an application pre-registered in the monitoring list is activated on the computing device; and a step of automatically acquiring one or more of the screen image and text from the computing device in response to the activation of the pre-registered application, wherein the step of automatically acquiring one or more of the screen image and text from the computing device includes: a step of determining an application identifier for acquiring one or more of the screen image and text; and a step of determining whether direct acquisition of text containing the user's work content is possible from the computing device. A method for assisting with business using an AI agent, comprising: a step of calling an operating system capture interface of the computing device to acquire a screen image when text containing the user's work content cannot be directly acquired from the computing device; a step of creating an acquired content object using one or more of the screen image and text acquired from the computing device; and a step of assigning the application identifier as metadata to the acquired content object. Claim 2 delete Claim 3 A method for assisting work using an AI agent, wherein the step of generating analysis information comprises: a step of setting an output format of the large-scale language model using feature values ​​extracted from the template information; and a step of inputting one or more of the screen image and text into the large-scale language model having the output format set. Claim 4 A method for assisting business using an AI agent according to claim 1, further comprising, prior to the step of generating the analysis information, a step of receiving a user's selection input regarding a template information to be used for the business result data among a plurality of template information pre-stored in the computing device or an external device capable of communicating with the computing device. Claim 5 A method for assisting work using an AI agent according to claim 1, further comprising the step of displaying the work result data on the computing device through a user interface that allows the user to modify the work result data. Claim 6 A computer program stored on a computer-readable recording medium to perform a work assistance method using an AI agent according to any one of claims 1 and 3 through 5, combined with hardware. Claim 7 A work assistance system using an AI agent executed by a computing device, comprising: a template input unit configured to receive template information defining the format of a work result; a work content acquisition unit configured to acquire one or more of a screen image and text of the computing device containing the user's work content in conjunction with the AI ​​agent; and an analysis information generation unit configured to generate analysis information about a work by using a feature value extracted from the template information and one or more of the screen image and text as input values ​​for a pre-trained large-scale language model. A work assistance system using an AI agent, comprising a result processing unit configured to process the analysis information into work result data corresponding to the template information, wherein the work content acquisition unit registers an application monitoring list and detects that an application previously registered in the monitoring list is activated on the computing device, and in response to the activation of the previously registered application, automatically acquires one or more of the screen image and text from the computing device, determines an application identifier for acquiring one or more of the screen image and text, determines whether text containing the user's work content can be directly acquired from the computing device, and if text containing the user's work content cannot be directly acquired from the computing device, calls the operating system capture interface of the computing device to acquire the screen image, creates an acquired content object using one or more of the screen image and text acquired from the computing device, and is further configured to automatically acquire one or more of the screen image and text by assigning the application identifier as metadata to the acquired content object. Claim 8 In claim 7, the above-mentioned work content acquisition unit is a work assistance system using an AI agent further configured to automatically acquire one or more of the screen image and text from the computing device in response to a preset application being activated on the computing device. Claim 9 In claim 7, the analysis information generation unit is further configured to set the output format of the large-scale language model using feature values ​​extracted from the template information, and to generate the analysis information by inputting one or more of the screen image and text into the large-scale language model with the set output format. This is a business assistance system using an AI agent. Claim 10 In claim 7, the above template input unit is further configured to receive a user's selection input regarding template information to be used for business result data among a plurality of template information stored in advance before the analysis information is generated, in a business assistance system using an AI agent. Claim 11 A work assistance system using an AI agent according to claim 7, further comprising a user interface module configured to display the above-mentioned work result data on the computing device and provide it for the user to modify.

Images