AI-based automatic outdoor business permit verification system

Abstract

The present invention relates to an AI-based automatic determination of eligibility for outdoor business operations and a licensing verification system, comprising: a user terminal equipped with a display unit and a user input unit; and a server connected to the user terminal via a wired or wireless network, wherein the server receives location information, drawing files, a building register, and business operator information of the business subject to licensing from the user terminal.

Description

Technology Field

[0001] The present invention relates to an AI-based automatic determination of eligible outdoor business targets and a permit verification system that achieves digital innovation to simultaneously increase administrative efficiency and citizen satisfaction. Background Technology

[0002] Traditional licensing administrative procedures were structured such that multiple laws—including those concerning food hygiene, construction, environment, and traffic—operated separately within distinct administrative systems. Generally, applicants had to manually verify the location, use, and area of ​​a business by personally examining documents such as building registers and land use plan confirmations, and then consult separate relevant departments to determine whether operation was permitted. During this process, cases frequently occurred where business registrations were filed without prior knowledge that the primary use of a building was not a neighborhood living facility; consequently, this resulted in administrative delays requiring the additional process of changing the building's use.

[0003] Furthermore, the verification of wastewater generation volumes or septic tank capacities by business type has mostly been performed by personnel directly inputting figures into Excel spreadsheets. However, as information on building gross floor area, number of floors, and structural details is scattered across multiple documents, problems have arisen where calculated wastewater generation volumes are underestimated or overestimated compared to the actual figures due to input errors or omissions in unit conversions. Consequently, consistency in the calculation of amounts has not been ensured during the polluter pays stage, and disputes frequently occur during the permit process due to differences in interpretation between applicants and administrative agencies.

[0004] Furthermore, the existing administrative system lacked the capability to automatically link and verify data regarding building uses or wastewater for permit applications, requiring staff to manually refer to legal codes or standard tables. Consequently, this led to prolonged permit processing times and presented a structural limitation in that it was difficult for practitioners handling a large volume of civil complaints to maintain consistent application of regulations. The problem to be solved

[0005] To solve the aforementioned problems, the present invention provides an AI-based automatic determination of eligibility for outdoor business operations and permit verification system that automatically identifies the unique number, main use, gross floor area, number of floors, structure, floor area ratio, site area, and regional / zone information of a building from a building register. Accordingly, without the need for a person in charge to directly input document figures or compare them with laws, the system can immediately determine whether the building falls under Class 1 or Class 2 neighborhood living facilities by calling up pre-stored standard data for building use classification.

[0006] In addition, by calling up pre-set standard data for wastewater generation based on the building's total floor area, number of floors, and structural information, and automatically calculating the wastewater generation volume (㎥ / day) according to a calculation formula, calculation errors caused by conventional manual input are fundamentally prevented. Through this, the calculation of wastewater volume and septic tank capacity during the licensing process is consistently performed based on objective figures, and the fairness in the calculation of polluter pays fees and the speed of administrative processing are simultaneously ensured.

[0007] In other words, the present invention provides an AI-based automatic determination of eligibility for outdoor business operations and permit verification system that integrates the automatic identification of building register information and the automatic calculation of wastewater generation volume, thereby resolving issues such as permit delays, discrepancies in legal interpretation, and calculation errors caused by manual calculation that occurred in conventional technology, and significantly improving the reliability and efficiency of permit administration. means of solving the problem

[0008] In an AI-based automatic determination of outdoor business eligibility targets and licensing verification system according to one embodiment of the present invention, a user terminal equipped with a display unit and a user input unit; and a server connected to the user terminal via a wired or wireless network; wherein the server, upon receiving building location information, drawing files, building register, business operator information, and outdoor business pre-confirmation checklist data entered by the user, converts the received data into structured input data through a prompt generation module that converts the received data into an LLM input prompt, and is configured to determine whether the building is eligible for outdoor business by querying the basis information linked to the building administration system (DB) and the law and ordinance system (DB) in a RAG manner using an LLM-based judgment module, and the server is configured to automatically determine whether there is an obligation to subscribe to disaster liability insurance based on the location, number of floors, and area information of the establishment, and to calculate the normalization results of the gross floor area and usage standards necessary for calculating the causer pays fee and whether the use is permissible for food service businesses based on building register data, and the server is configured to provide basic administrative requirements including septic tank capacity calculation standards (whether there is a shortage compared to the required capacity), the scope of business reporting per industry code, setback standards for the site and building line, and whether there are restrictions on separation distance and landscaping area according to local government ordinances. It is characterized by collecting and normalizing judgment data and adding it to the structured input data, and configuring the basic administrative requirement judgment data to be utilized as a component of the primary judgment data and LLM input prompt generated in a subsequent stage.

[0009] The above server includes a communication unit; a storage unit; and a control unit; wherein the control unit uses the location information to query the building register information in the building administration system (DB), identifies responses to multiple items included in the outdoor business pre-confirmation checklist to generate primary determination data regarding whether the establishment is eligible for outdoor business or subject to restrictions, identifies the suitability of the main use in the building register (whether it is a neighborhood living facility), the possibility of business registration according to the region / district, the method of contiguous access to the outdoor business area (whether there is direct access on the same floor, whether it is contiguous within the site), whether evidence proving the right to use the outdoor business area is satisfied, whether facility standards are satisfied, the result of septic tank capacity calculation (whether it is insufficient compared to the required capacity), whether it is subject to the polluter pays principle (determination by type of new construction, expansion, or change of use), and whether it is subject to mandatory disaster liability insurance based on whether the area is 100㎡ or more, combines this with the primary determination data, generates an LLM prompt with the primary determination data as input to query the LLM-based judgment module, and the judgment module [reviews] the building administration DB and laws / ordinances The control unit controls the generation of a judgment result among “outdoor business possible,” “outdoor business restricted,” and “additional review required” based on the provisions of the system (DB) and provides it to the user terminal. The control unit sets importance weights for each item within the LLM prompt so that the primary judgment data is utilized for drawing-based analysis, identification of restricted areas under other laws, and determination of whether ordinance standards are satisfied. It also configures a judgment path (branch) that the LLM can select according to a combination of judgment items related to adjacent types, usage rights, septic tanks, insurance, and charges, and controls the configuration of the judgment path so that it is applied as an essential input condition for the subsequent drawing analysis stage and restricted area mapping stage.

[0010] The control unit extracts object information, including the location, boundary lines, and entry / exit routes of the outdoor business area contained in the drawing file, using computer vision (CV) and OCR analysis technologies; integrates the extracted drawing-based information with the building use, gross floor area, structure, and site area stored in the building administration system (DB) into the input prompt of the LLM judgment module; performs drawing-based area calculation, determines whether routes are separated, checks for obstructions in emergency passages, checks whether restricted areas under other laws regarding public open spaces, evacuation zones, road occupancy zones, and mandatory parking areas are included, and identifies conditions requiring confirmation by departmental personnel; and controls the transmission of warning data to the user terminal based on the results identified by the judgment module, which displays the possible outdoor business area and scope on the drawing when permitted, and includes the name of the violating statute, article number, grounds, and contact information of the relevant department when restricted. The control unit also controls the direct accessibility between the entrance and the outdoor business area on the drawing, the possibility of inter-floor movement and contiguousness within the site, and other use spaces regarding common areas, corridors, and parking lots. It is characterized by determining whether intervention is necessary based on pre-established objective criteria, quantifying the possibility of inclusion in a restricted area by calculating the overlap ratio with restricted areas under other laws and the center coordinate error range, and classifying items requiring departmental review corresponding to each restriction condition to generate a structured restriction condition dataset that can be transmitted along with the above judgment results.

[0011] The control unit identifies whether the location of the building or outdoor business premises falls within any of the following: public open space, open space within the site, landscaped area, evacuation passage, road occupancy area, or mandatory parking area, by matching the location information, floor plan, and layout data of the building with a previously stored database of restriction conditions under other laws; if it falls within the restriction area, it generates warning data containing the name of the relevant statute, article number, name of the responsible department, and contact information for each item where a risk of violation exists; transmits the warning data to the electronic signature module of each relevant department to collect signatures; and controls the issuance of a permit verification result report to the user terminal once signatures from all relevant departments are completed.

[0012] The control unit calculates the statistical similarity and pattern matching rate for each object by comparing the location coordinates and attribute data of an object extracted from the drawing file with first historical data corresponding to the same industry, similar buildings, and similar districts among the log data of past permit verification results accumulated in the storage unit, and if the similarity and pattern matching rates are each below a preset threshold, the building is classified as a subject for in-depth review, and if the similarity and pattern matching rates exceed the threshold, the verification procedure is controlled to continue. Effects of the invention

[0013] We provide an AI-based automatic determination of eligibility for outdoor business operations and permit verification system that automatically identifies the building's unique number, main use, gross floor area, number of floors, structure, floor area ratio, site area, and regional / zone information from the building register. Accordingly, without the need for a person in charge to directly input document figures or compare them with laws, the system can immediately determine whether the relevant building falls under Class 1 or Class 2 neighborhood living facilities by calling up pre-stored standard data for building use classification.

[0014] In addition, by calling up pre-set standard data for wastewater generation based on the building's total floor area, number of floors, and structural information, and automatically calculating the wastewater generation volume (㎥ / day) according to a calculation formula, calculation errors caused by conventional manual input are fundamentally prevented. Through this, the calculation of wastewater volume and septic tank capacity during the licensing process is consistently performed based on objective figures, and the fairness in the calculation of polluter pays fees and the speed of administrative processing are simultaneously ensured.

[0015] In other words, by integrating the automatic identification of building register information and the automatic calculation of wastewater generation volume, the present invention can provide an AI-based automatic determination of outdoor business eligibility and permit verification system that resolves problems such as permit delays, discrepancies in legal interpretation, and calculation errors caused by manual calculation that occurred in conventional technology, and has the effect of significantly improving the reliability and efficiency of permit administration. Brief explanation of the drawing

[0016] FIG. 1 is a diagram illustrating the schematic configuration of an AI-based automatic determination of outdoor business eligibility targets and licensing verification system according to one embodiment of the present invention. FIG. 2 is a diagram illustrating the specific configuration of an AI-based automatic determination of outdoor business eligibility targets and licensing verification system according to one embodiment of the present invention. FIG. 3 is a flowchart illustrating the operation of an AI-based automatic determination of outdoor business eligibility targets and licensing verification system according to one embodiment of the present invention. FIG. 4 is a drawing illustrating an example of a current status diagram of a building for which an outdoor business consultation is requested in an AI-based automatic determination of potential outdoor business targets and a permit verification system according to an embodiment of the present invention. FIG. 5 is a drawing illustrating an example of an external photograph of a building for which an outdoor business consultation was requested in an AI-based automatic determination of potential outdoor business targets and a permit verification system according to an embodiment of the present invention. Specific details for implementing the invention

[0017] Embodiments of the present invention will be described in detail below with reference to the accompanying drawings. In the drawings, identical reference numbers or symbols refer to components that perform substantially the same function, and the size of each component in the drawings may be exaggerated for clarity and convenience of explanation. However, the technical concept of the present invention and its core components and operations are not limited only to the components or operations described in the following embodiments. In describing the present invention, if it is determined that a detailed description of known technologies or components related to the present invention may unnecessarily obscure the essence of the present invention, such detailed description will be omitted.

[0018] In embodiments of the present invention, terms including ordinal numbers, such as first, second, etc., are used solely for the purpose of distinguishing one component from another, and singular expressions include plural expressions unless the context clearly indicates otherwise. Furthermore, in embodiments of the present invention, terms such as 'composed of,' 'include,' 'have,' etc., should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof. Additionally, in embodiments of the present invention, 'module' or 'part' performs at least one function or operation and may be implemented in hardware or software, or as a combination of hardware and software, or may be integrated into at least one module and implemented as at least one processor. Furthermore, in embodiments of the present invention, 'at least one' among a plurality of elements refers not only to all of the plurality of elements but also to each individual element excluding the remainder or all combinations thereof. Additionally, "configured to" may be used interchangeably with, depending on the context, for example, "suitable for," "having the capacity to," "designed to," "adapted to," "made to," or "capable of." "Configured to" does not necessarily mean that it is "specifically designed to" in hardware. Instead, in some situations, the expression "device configured to" may mean that the device is "capable of" doing so in conjunction with other devices or components.For example, the phrase “a processor configured (or set) to perform A, B, and C” may mean a dedicated processor for performing said operations (e.g., an embedded processor), or a generic-purpose processor (e.g., a CPU or an application processor) capable of performing said operations by executing one or more software programs stored in a memory device.

[0019] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the drawings. This description is intended to be detailed enough for a person skilled in the art to easily practice the invention, and it should be noted that the technical scope and concept of the present invention are not limited thereby.

[0020] FIG. 1 is a diagram illustrating the schematic configuration of an AI-based automatic determination of outdoor business feasibility and licensing verification system according to an embodiment of the present invention; FIG. 2 is a diagram illustrating the specific configuration of an AI-based automatic determination of outdoor business feasibility and licensing verification system according to an embodiment of the present invention; FIG. 3 is a diagram illustrating the operation of an AI-based automatic determination of outdoor business feasibility and licensing verification system according to an embodiment of the present invention; FIG. 4 is a diagram illustrating an example of a current status diagram of a building requesting consultation for outdoor business in an AI-based automatic determination of outdoor business feasibility and licensing verification system according to an embodiment of the present invention; and FIG. 5 is a diagram illustrating an example of an external photograph of a building requesting consultation for outdoor business in an AI-based automatic determination of outdoor business feasibility and licensing verification system according to an embodiment of the present invention.

[0021] Referring to FIGS. 1 to 5, an AI-based automatic determination of outdoor business eligibility and licensing verification system according to one embodiment of the present invention comprises: a user terminal equipped with a display unit and a user input unit; and a server connected to the user terminal via a wired or wireless network, wherein the server receives location information, drawing files, building registers, and business operator information of the business subject to licensing from the user terminal.

[0022] Meanwhile, according to one embodiment of the present invention, there may be multiple user terminals, and it goes without saying that the multiple user terminals may be connected to a server simultaneously or separately.

[0023] A user terminal (100) according to one embodiment of the present invention includes, for example, a personal computer, a server computer, a handheld or laptop device, a mobile device (mobile phone, PDA, media player, etc.), a multiprocessor system, a consumer electronic device, a mini computer, a mainframe computer, a distributed computing environment including any of the aforementioned systems or devices, and an edge computing environment in which data is processed at the edge where data is generated rather than at a central server, and the configuration is not limited only to what is described.

[0024] The user terminal (100) may include at least one processor and memory. Here, the processor may include, for example, a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), and may have multiple cores.

[0025] The memory may be volatile memory (e.g., RAM, etc.), non-volatile memory (e.g., ROM, flash memory, etc.), or a combination thereof. Additionally, the user terminal (100) may include additional storage. The storage includes, but is not limited to, magnetic storage, optical storage, etc. The storage may store computer-readable instructions for implementing one or more embodiments disclosed herein, and may also store other computer-readable instructions for implementing an operating system, an application program, etc. Computer-readable instructions stored in the storage may be loaded into memory to be executed by a processor.

[0026] Additionally, the user terminal (100) may include a user input unit (110) and an output device. The user input unit (110) may include, for example, a keyboard, a mouse, a pen, a voice input device, a touch input device, an infrared camera, a video input device, or any other input device. Additionally, the output device may include, for example, one or more displays, speakers, printers, or any other output devices. Additionally, the computing device may use an input device or an output device provided in another computing device as the user input unit (110) or output device. Additionally, the computing device may include a communication module that enables the computing device to communicate with another device. Here, the communication module may include a modem, a network interface card (NIC), an integrated network interface, a radio frequency transmitter / receiver, an infrared port, a USB connection, or other interfaces for connecting the computing device to another computing device. The communication module may include a wired connection or a wireless connection.

[0027] Each component of the user terminal (100) may be connected by various interconnections such as a bus (e.g., peripheral component interconnection (PCI), USB, firmware (IEEE 1394), optical bus structure, etc.) or may be interconnected by a network. Terms such as “component,” “system,” etc. used in this specification generally refer to computer-related entities that are hardware, a combination of hardware and software, software, or running software.

[0028] A user terminal (100) according to one embodiment of the present invention may include a display unit (120), and the method of implementing the display of the display unit (120) is not limited and may be implemented in various display methods such as liquid crystal, plasma, light-emitting diode, organic light-emitting diode, surface-conduction electron-emitter, carbon nanotube, nanocrystal, etc. In the case of a liquid crystal method, the display unit (120) includes a liquid crystal display panel, a backlight unit that supplies light to the liquid crystal display panel, and a panel driving unit that drives the liquid crystal display panel. Meanwhile, the display unit (120) may be implemented as an OLED panel, which is a self-emissive element, without a backlight unit.

[0029] Artificial intelligence (AI) according to one embodiment of the present invention refers to the entirety of information processing technology configured to mimic human learning, reasoning, judgment, and prediction functions using computing resources. Such AI is not limited to a single algorithm and can be implemented in various forms depending on the type of data, processing purpose, and learning method. For example, different learning structures such as machine learning, deep neural networks, reinforcement learning, and natural language processing can be applied, and these can be operated individually or in combination.

[0030] AI can generally be composed of a preprocessing unit that collects input data, a learning model that interprets the collected data using specific rules or functions, and an inference engine that makes judgments on new inputs based on the learned results. The preprocessing unit transforms raw data acquired from sensors, networks, user terminals, databases, etc., into a form that the model can process through procedures such as normalization, quantification, or filtering; in this process, it can perform operations such as outlier detection, missing value correction, unit conversion, and data sampling.

[0031] The learning model updates parameters to minimize the error between input data and target values ​​based on a statistical probability function, a loss function, or an objective function, and the learning method may be one or more of supervised learning, unsupervised learning, semi-supervised learning, self-supervised learning, or reinforcement learning. For example, in the case of supervised learning, backpropagation is performed by calculating the difference between the predicted value and the actual value for labeled data pairs (x, y), and as a result, the weights and biases of each layer are adjusted.

[0032] The inference engine performs predictions or judgments on unknown input data based on parameters accumulated by the learning model in the storage unit, and its output values ​​can be provided in various forms such as probability, score, level, or label. Furthermore, the inference result may be configured not to be limited simply to a confirmed value, but to output uncertainty or confidence in parallel, thereby enabling a subsequent judgment module to perform correction or reinference.

[0033] The AI ​​used in one embodiment of the present invention may be implemented in a centralized server environment, a cloud computing environment, or a distributed structure based on edge devices, and each node may learn independently depending on the characteristics of the data, or exchange information with one another using federated learning or knowledge distillation methods. In this case, the structure of the learned model may be formed as a multilayer perceptron, a convolutional neural network (CNN), a recurrent neural network (RNN), a transformer, a graph neural network (GNN), or a combination thereof.

[0034] Furthermore, AI is not limited to statistical models that simply derive correlations between data, but can also be realized as a self-adaptive structure that estimates causal relationships between multiple input variables or readjusts model parameters in real time according to changes in the external environment. To improve prediction accuracy, such AI can use ensemble learning, weighted averaging, or Bayesian inference in combination.

[0035] Accordingly, the AI ​​(artificial intelligence) according to one embodiment of the present invention is not limited by a specific algorithm, formula, or data structure, and includes any information processing system capable of a computer identifying rules, patterns, or relationships from data and performing judgment, inference, or prediction regarding new inputs based thereon. This may be implemented in the form of physical hardware, virtualized software, or a combination of both, and the specific methodology for learning and inference may be varied in many ways depending on the application target, data characteristics, and the designer's purpose.

[0036] A licensing verification according to one embodiment of the present invention refers to a series of procedures for determining whether a specific facility, equipment, building, product, or service satisfies the requirements stipulated by relevant laws, technical standards, administrative regulations, or industry standards. Such licensing verification is not limited to a simple administrative approval process and includes the determination of conformity regarding various factors, such as physical structures, digital systems, environmental factors, or human requirements.

[0037] Permit verification generally involves multiple procedures, including application, review, evaluation, and approval stages. In the application stage, basic information, design drawings, technical data, experimental results, and test reports regarding the subject of verification are collected. In the review stage, a procedure is performed to compare and contrast the relevant data with provisions of applicable laws, guidelines, or technical standards. In the evaluation stage, quantitative and qualitative assessments are conducted across multiple categories, such as structural safety, hygiene, environmental impact, and fire and evacuation standards. Finally, in the approval stage, the evaluation results are recorded and preserved, and the outcome is issued in the form of written or electronic approval by the administrative agency.

[0038] The aforementioned licensing verification can be classified according to the performing entity into automated verification by the administrative agency's internal system, manual verification by 담당 officials, or technical verification by external specialized institutions. In the case of automated verification, the compliance of conditions is determined by comparing input values ​​with standard values ​​based on a computerized database, whereas in manual verification, the person in charge directly verifies compliance by referring to relevant laws and regulations. External technical verification may be performed in the form of certification bodies or testing laboratories providing objective data through physical and chemical measurements, experiments, simulations, etc.

[0039] The reference data applied to licensing verification may be derived from laws, public notices, ordinances, technical standards, international standards (e.g., ISO, IEC), or industry-specific guidelines. For instance, in the construction sector, structural safety standards, seismic performance standards, and evacuation and fire protection structural standards may be included as verification items; in the environmental sector, wastewater discharge standards, septic tank capacity standards, and permissible air pollution concentrations may be established as verification elements. Furthermore, in the food and pharmaceutical sectors, hygiene management standards, permissible limits for raw materials and additives, and facility hygiene standards may be included. Such reference data may be updated periodically, and the verification system may be configured to maintain the reliability of verification results by reflecting the latest reference values.

[0040] The execution process of the aforementioned licensing verification may include steps for checking the consistency of input data, searching and retrieving reference data, performing comparison operations, and storing and outputting judgment results. The consistency check is a procedure to verify whether the format, units, coordinate systems, etc., of the input documents or drawings comply with the specifications, and the comparison operation is a process of mathematically comparing the attribute values ​​of the data under verification with the allowable range of the reference data. Judgment results may be classified into multiple grades, such as conformity, conditional conformity, or non-conformity, and each result may be used as a basis for administrative judgment or approval decisions in subsequent procedures.

[0041] In addition, licensing verification can be performed in administrative information systems, cloud servers, or standalone local server environments, and in cases where multiple agencies or departments participate, an integration module for data exchange may be included. Such integration modules serve to coordinate differences in verification standards by law or agency, or to prevent duplicate verification.

[0042] The licensing verification according to one embodiment of the present invention is not limited by specific subjects, procedures, standards, or execution environments, and includes all forms of procedural verification activities capable of determining suitability by judging and identifying attribute information of a specific object according to certain standards or rules. Accordingly, it can be applied to various application fields such as administrative, technical, environmental, industrial, or data-based verification, and the specific algorithms, data structures, and processing methods for implementing this can be modified in various ways depending on the designer's purpose and the characteristics of the verification subject.

[0043] An establishment subject to licensing according to one embodiment of the present invention refers to any type of workplace, facility, or place that requires prior permission, notification, registration, or certification to carry out activities such as business operations, installation, operation, construction, manufacturing, or service provision in accordance with relevant laws, rules, ordinances, or administrative guidelines of the state or local government. Different legal applications and verification standards may be applied to such establishments subject to licensing depending on their business type, location, use, facilities, personnel size, etc., and include not only tangible establishments based on physical space but also non-face-to-face workplaces that utilize non-physical data processing facilities as their primary means.

[0044] Businesses subject to licensing can generally be classified into commercial facilities, production facilities, sanitation facilities, medical facilities, educational facilities, cultural facilities, and public facilities; each facility is defined according to industry classification codes stipulated in relevant laws, the Standard Industrial Classification, or permitted uses under local government ordinances. For example, food-related businesses may be subject to business permits or notifications under the Food Sanitation Act, and if specific structures or facilities installed in buildings are operated, licensing procedures under other laws, such as the Building Act or the Firefighting Facilities Act, must be followed concurrently. Furthermore, various business sectors, including manufacturing, environmental management, medical institutions, accommodation facilities, transportation businesses, and information and communication service providers, may fall under the category of businesses subject to licensing, each with distinct licensing standards.

[0045] The aforementioned establishments subject to licensing may be subdivided into new establishment, modification, relocation, consolidation, and division types depending on their operational form. The new establishment type refers to the initial licensing or notification procedures accompanying the opening of a new business location, while the modification type includes administrative procedures regarding changes to the structure, area, use, name, representative, etc., of an existing establishment. The relocation type refers to cases where new licensing verification is required due to the relocation of an establishment of the same business type to a different site or building; the consolidation type refers to cases where multiple establishments are integrated and managed as a single business location; and the division type refers to cases where a single business location is separated into multiple independent establishments.

[0046] The physical components of establishments subject to licensing may include the main building structure, ancillary facilities, sanitary facilities, ventilation facilities, wastewater treatment facilities, electrical and gas facilities, heating and cooling facilities, evacuation routes, and parking facilities; these components may be added or removed variably depending on the characteristics of the business type. For example, restaurants may include kitchens, dining halls, washing rooms, water supply and drainage facilities, ventilation ducts, and septic tanks, while medical institutions may include ancillary spaces such as examination rooms, operating rooms, medical waste storage areas, drug storage rooms, and computer records rooms. These components may be used as primary evaluation targets during licensing verification, and it will be determined whether the location, structure, area, use, and connectivity of each component comply with the licensing standards.

[0047] In addition, establishments subject to licensing may be classified into sole proprietorships, corporations, public institutions, non-profit organizations, or temporary facilities depending on the legal nature of the operating entity. In this case, even for the same business type, the licensing requirements may vary depending on the legal status of the operating entity, the size of capital, the number of employees, or the location of the workplace. For example, even for the same food service business, a corporation may be subject to additional obligations, such as designating a hygiene manager, subscribing to insurance, and completing training, in addition to facility standards.

[0048] Information regarding businesses subject to licensing can consist of various public and private data, such as building registers, land use plan confirmations, business registration certificates, industry classification tables, design drawings, technical data, test reports, and certifications. This information can be used as input data for the licensing verification system and serves as a basis for comprehensively assessing the business's location, use, structure, and environmental impact.

[0049] According to one embodiment of the present invention, establishments subject to licensing are not limited to specific business types or facilities, but include all workplaces or installation targets that require prior verification, permission, notification, or registration procedures in any form under relevant laws and regulations. Accordingly, the scope may be flexibly expanded depending on the type of business activity, physical scale, legal requirements, or technical nature, and the system designer may freely set detailed items, attribute data, and verification items for establishments subject to licensing according to the characteristics of the application field.

[0050] A server (200) according to one embodiment of the present invention may include a communication unit (210), a storage unit (220), and a control unit (230).

[0051] A communication unit (210) according to one embodiment of the present invention can communicate with a user terminal (100) or other external electronic devices, etc., via wired or wireless communication methods. Therefore, in addition to a connection unit including a connector or terminal for wired connection, it can be implemented in various other communication methods. For example, it can be configured to perform one or more of Wi-Fi, Bluetooth, Zigbee, infrared communication, Radio Control, Ultra-Wide Band (UWM), Wireless USB, and Near Field Communication (NFC). The communication unit (210) may include communication modules such as Bluetooth Low Energy (BLE), Serial Port Profile (SPP), Wi-Fi Direct, infrared communication, Zigbee, and Near Field Communication (NFC). Additionally, the communication unit (210) can be implemented in the form of a device, a software module, a circuit, a chip, etc.

[0052] A communication unit (210) according to one embodiment of the present invention may include various communication modules as described above, and may include an IoT communication module having an IoT network for each carrier. An IoT communication module may refer to any IoT communication network in which a plurality of objects having separate communication units (210) are connected through a network to enable services based on various platforms. By using such an IoT communication module, a smoother communication network can be provided within a set area.

[0053] A storage unit (220) according to one embodiment of the present invention may receive information from a user terminal (100) or from an external search platform through a communication unit (210) and store it, and may receive text information and image information included in a plurality of web pages received by a control unit (230) and store them. The storage unit (220) may store various data according to the processing and control of the control unit (230) described later. The storage unit (220) may be accessed by the control unit (230) to perform reading, recording, modification, deletion, updating, etc. of data. The storage unit (220) may include non-volatile memory such as flash memory, hard disk drive, and SSD (solid-state drive) so as to preserve data regardless of whether system power is provided to the server (200). Additionally, the storage unit (220) may include volatile memory, such as a buffer or RAM, for temporarily loading data processed by the control unit (230).

[0054] A control unit (230) according to one embodiment of the present invention can perform control for the operation of various components of a server (200). The control unit (230) may include a control program (or instruction) that enables such control operation, an inactive memory in which the control program is installed, a volatile memory in which at least a portion of the installed control program is loaded, and at least one processor or CPU (Central Processing Unit) that executes the loaded control program. In addition, such a control program may be stored in an external electronic device other than the server (200).

[0055] The control program may include program(s) implemented in at least one form among BIOS, device driver, operating system, firmware, platform, and application. In one embodiment, the application may be pre-installed or stored in the server (200) at the time of manufacturing the server (200), or may be installed in the server (200) based on the received data by receiving data of the application from an external source when it is used later. The data of the application may be downloaded to the server (200) from an external server, such as an application market, for example, a platform according to the present invention, but is not limited thereto. Meanwhile, the control unit (230) may be implemented in the form of a device, S / W module, circuit, chip, etc., or a combination thereof.

[0056] According to one embodiment of the present invention, when the server receives building location information, drawing files, building registers, business operator information, and outdoor business pre-confirmation checklist data entered by a user, it converts the received data into structured input data through a prompt generation module that converts the data into an LLM input prompt.

[0057] When the server according to one embodiment of the present invention receives building location information, drawing files, building registers, business operator information, and outdoor business pre-confirmation checklist data input from a user terminal, the control unit first identifies the format of each input data. For example, location information may be received in the form of a string or JSON consisting of latitude and longitude coordinates, and drawing files may be received as PDF, images (JPG, PNG), or CAD-based vector data. In addition, the building register and business operator information may be received in XML, JSON, or CSV formats depending on the integration with the administrative system API, and checklist data is stored in a Key-Value structure in which selected values ​​for each item are mapped. To convert the received data of individual formats into an internal common data structure, the control unit calls a data preprocessing module to perform string normalization, conversion of coordinate data into numeric types, verification of file format header information, and verification of missing fields.

[0058] The control unit transmits the preprocessed input data to the prompt generation module, and the prompt generation module integrates data items into a natural language-based structured sentence or JSON structure to match the input schema required by the LLM-based judgment module. For example, the prompt generation module converts “building location information” into a text block in the form of “location: {lat: ###, lon: ###}”, rearranges “building register information” into a structure with clearly distinguished field names such as “building_register: {use: 'Neighborhood Living Facility', area: ###, floors: ###}”, and configures the checklist to be displayed in the form of a list of “checklist: [Item 1: Response Value 1, Item 2: Response Value 2, …]”. At this time, the prompt generation module performs a procedure to generate a hash value to verify whether the original value and the converted value of each item match, for example, by calculating a SHA-256-based hash value hash(S) for the input string S and storing it together with the converted data to prevent subsequent conversion errors.

[0059] The aforementioned prompt generation module is also configured to compare whether the input coordinates match the site location code and administrative district code included in the building register for cross-verification between location information and the building register, and to automatically include a status flag such as “location_verification: mismatch_detected” within the LLM input prompt if a discrepancy is detected. This supports the LLM judgment module in performing inference that considers even the relationships between input data.

[0060] In addition, the prompt generation module performs basic verification to ensure that text, object coordinates, and area values ​​extracted from drawing files do not logically contradict each other with the number of floors, gross floor area, and structural information from the building register. For example, if the outdoor business area value A_draw extracted from the drawing is greater than the site area A_land in the building register, text in the form of a conditional statement, such as “area_warning: true, A_draw > A_land”, is automatically inserted into the prompt. This information enables the LLM to recognize potential errors in subsequent judgments.

[0061] The prompt ultimately generated by the prompt generation module is not in the form of a single paragraph, but is composed of partitioned areas such as the “Input Data Summary Block,” “Architectural Administration DB Query Target Field Block,” “Checklist Response Block,” and “Drawing-based Detection Information Block,” and each area consists of explicit tags (e.g., <location>, <REGISTER_DATA>, <checklist>,<DRAWING_INFO> It is separated by ) allowing the LLM to recognize each data group independently during context analysis. This structure helps improve accuracy by enabling the model to focus on analyzing only specific fields, even when prompt lengths become long.

[0062] The prompt generation module is configured to rearrange the order of items within the prompt according to a fixed rule, wherein the rule follows the sequence of "Location Information → Building Register Summary → Drawing Information Summary → Checklist Response → Additional Verification Tag." The prompt configured in this way is stored as a standard format for input data to be used by the LLM-based judgment module, and a hash value in the form H = hash(P) is generated for the entire prompt string P converted at the time of storage to record it, thereby enabling verification of whether it has been modified compared to the original input data.

[0063] According to one embodiment of the present invention, the server is configured such that an LLM-based judgment module queries supporting information linked to a building administration system (DB) and a law and ordinance system (DB) in a RAG manner to determine whether the building is eligible for outdoor business operations. Accordingly, the judgment module first analyzes structured input data transmitted from a prompt generation module, converts elements such as the building's use, location, site structure, area, and checklist response value—which are the core of the query in the input data—into a natural language-based query, and the control unit converts the query into a search request format for the building administration system (DB) and the law and ordinance system (DB) and transmits it to the information retrieval unit.

[0064] The information retrieval unit matches documents within the administrative DB corresponding to the input query (e.g., building laws, ordinances, guidelines, land information, usage classification tables, facility restriction regulations, etc.) with a vector store that stores them in an indexed form. Each document is divided into sentence or item units in advance, converted into an embedding vector, and stored. At this time, the embedding is converted into a fixed-dimensional vector by an embedding function of the form "embedding(x)". The judgment module generates embedding(q) for the input query q and identifies the most semantically relevant supporting document by calculating the cosine similarity cos(embedding(q), embedding(d)) between the corresponding vector and the document in the vector store. The above similarity calculation can be expressed in Python as similarity = dot(q, d) / (sqrt(sum(q^2)) × sqrt(sum(d^2))).

[0065] The aforementioned judgment module selects the top k source documents with the highest similarity, organizes them into 'search result context blocks,' and combines these blocks with an input prompt to transmit them to the LLM. In this process, the RAG method provides "input prompt + searched relevant statutes / ordinance provisions + building administration information" as a single context, thereby inducing the model to perform inference based on legal grounds or administrative information rather than simple summarization. For example, if the searched provisions include the statement "Areas restricted for outdoor business operations correspond to evacuation routes, landscaping areas, and public open spaces," the judgment module includes that provision as is in the context, allowing the LLM to utilize it as the basis for judgment.

[0066] The LLM-based judgment module receives the aforementioned combined prompt and derives, through natural language inference, whether the conditions of the relevant building fall under either permissible or restrictive requirements. In this process, the judgment module compares the semantic patterns of keywords within the searched provisions (e.g., "prohibited," "permitted," "exception," "area exceeded," "road access conditions," "inclusion of open space within the site") with factual data extracted from the input data (e.g., area values, location coordinates, drawing-based boundary information). Additionally, the judgment module operates to comprehensively determine whether the relevant conditions correspond to grounds for restriction under laws or ordinances by also considering automatically generated status flags within the prompt (e.g., location discrepancy, area exceeded, etc.).

[0067] The judgment module ultimately generates at least one judgment result among "Outdoor business permitted," "Outdoor business restricted," and "Further review required," and this result is returned to the control unit in JSON or natural language format. The control unit automatically extracts the article number, title, and parent category of the source documents referenced by the LLM from the returned judgment result, organizes them into a source table, and configures this to be included in the verification result screen or PDF-based report transmitted to the user terminal. This operation clearly demonstrates that the judgment module's reasoning process is not arbitrary but is based on objective evidence retrieved from the actual database.

[0068] Furthermore, when generating judgment results, the judgment module utilizes simple comparison formulas to determine whether key items within the input data exceed standard conditions, fall short, or are inconsistent with ordinances. For instance, if an outdoor business area is identified as encroaching upon the public open space on the drawings, the judgment module internally sets a logical variable such as "exclusive_area_conflict = true." It is configured to analyze this in conjunction with supporting documents to arrive at a judgment of "restriction" or "further review." This comparison and setting process goes beyond simple rule-based judgments and combines with LLM inference to generate a more reliable final judgment.

[0069] According to one embodiment of the present invention, the server is configured to automatically determine whether there is an obligation to subscribe to disaster liability insurance based on the location, number of floors, and area information of an establishment, and to calculate the normalization results of the gross floor area and usage standards necessary for calculating the causer pays principle, as well as whether the use is permissible for food service businesses, based on building register data.

[0070] The server first converts the location information received from the user terminal into an administrative district code and queries the scope of disaster liability insurance coverage for each local government in the insurance standard table of the storage unit. Subsequently, using the number of floors and the business area of ​​the establishment as input values, it calculates whether insurance is mandatory by comparing conditional expressions notified by each local government, such as "mandatory enrollment for business areas of 100㎡ or more" and "mandatory enrollment for basement floors," in a Boolean form. For example, if area value A is greater than or equal to standard area B (A ≥ B) or if the number of floors C is included in the list of floors subject to mandatory coverage, the insurance_required variable is set to true, and this variable is subsequently used as a regulatory requirement item in the LLM input prompt.

[0071] In addition, the server parses the main use and secondary use codes, gross floor area, and floor area information of the building register retrieved from the building administration system (DB), and compares them with the list of permitted use codes for food service businesses in the Korean Standard Industrial Classification (KSIC) table. The list of permitted use codes includes multiple codes stored in advance, such as Class 1 and Class 2 neighborhood living facilities, and the server determines whether the use code in the building register matches the list through string-based matching or code number normalization methods. If the use codes do not directly match, the server calls the 'similar use code mapping table' stored in the storage unit to construct a two-stage determination result, such as "entertainment facility → food service business not allowed" or "sales facility → additional verification of possibility."

[0072] In addition, regarding the calculation of the developer pays fee, the server generates the burden_base_dataset by normalizing all attribute data, such as the gross floor area value from the building register, zoning (e.g., general commercial area, Class 2 general residential area), building setback status, and history of new construction, expansion, or change of use. The gross floor area value is normalized into a numeric type according to the unit conversion rule (M2 → ㎡), and the zoning value is mapped to the fee calculation intervals defined in the ordinances for each city, county, or district (e.g., Interval 1: 0–200㎡, Interval 2: 200–400㎡, etc.). The server compares the normalized gross floor area value R with the threshold range T of the ordinance interval to automatically identify intervals satisfying the conditions R ≥ T_min and R < T_max, and determines whether the corresponding interval is subject to fee calculation. During this process, the burden_flag variable is set to true or false so that it can be utilized as an essential element when the LLM judgment module searches for fee-related provisions thereafter.

[0073] The server generates structured regulatory requirement data that includes whether insurance is mandatory, whether the use is permissible for food service businesses, and whether a levy can be calculated. This data is stored at the field level (e.g., {insurance_required:true, food_service_allowed:false, burden_applicable:true}) and subsequently transmitted as is to the LLM prompt component during the judgment result generation stage. Since these data generation and processing procedures are all performed based on a predefined standard regulatory requirement model, a person skilled in the art can implement them in the same manner without separate estimation.

[0074] According to one embodiment of the present invention, the server collects and normalizes basic administrative requirement judgment data including standards for calculating septic tank capacity (whether it is insufficient compared to the required capacity), the range of business reporting possible by industry code, standards for setback of land and building lines, and whether there are restrictions on separation distance and landscaping area according to local government ordinances, and adds this to the structured input data, and is configured so that the basic administrative requirement judgment data is utilized as a component of the first judgment data and LLM input prompt generated thereafter.

[0075] The server first extracts the gross floor area, number of floors, and coordinate-based layout information from the building register and drawing files uploaded by the user, and compares this with the septic tank calculation standard table to determine whether there is a shortage relative to the required capacity. At this time, the required septic tank capacity is calculated by multiplying the wastewater generation rate per person by industry or the standard generation rate based on the number of seats by the standard value stored in the storage (required_capacity = base_rate × unit_count), and records whether required_capacity > capacity_installed as a Boolean value by comparing it with the current facility capacity (capacity_installed). This Boolean value is structured as the "Septic Tank Shortage Status" field.

[0076] Furthermore, the scope of business registration per industry code is determined based on the Standard Industrial Classification of Food Service Businesses and the list of permitted business types defined in each local government ordinance. The server determines whether a correspondence exists by comparing the primary and secondary use codes in the building register with the list of permitted business types using a string-code mapping method. For example, even if the use code corresponds to 'Neighborhood Living Facility,' some local government ordinances may restrict specific detailed business types; therefore, the server generates the allowed_use variable by querying the relevant local government's unique regulation table (local_regulation_table) and evaluating the possibility of business registration using the prescribed rule expression exactly.

[0077] Additionally, the server extracts the site boundary lines as vector data from the submitted site plan or building line drawing and compares the building line setback criteria with the local government ordinance standards. The required building line setback distance is stored as the setback_required value, and the server evaluates the condition by calculating the actual distance value (actual_setback_distance) between the nearest point of the building outline and the site boundary line (actual_setback_distance ≥ setback_required). The distance calculation is performed using the sqrt((x2 - x1)^2 + (y2 - y1)^2) method with the coordinates (x1,y1) and (x2,y2) on the drawing, and this value is normalized as is and recorded in the setback_compliant field.

[0078] In addition, restrictions on separation distance and landscaping area, which vary by local government, are determined according to standardized rule expressions (e.g., min_green_area_ratio, min_distance_from_boundary, etc.) stored in the ordinance metadata (regulation_metadata). The server normalizes the outdoor business area, site area, and green area calculated from the drawing file, and then evaluates condition expressions such as (green_area / lot_area) ≥ min_green_area_ratio and (distance_to_adjacent_lot ≥ min_distance_from_boundary), respectively. The results calculated in this way are stored as a structured data object called restriction_flags, and each field is recorded in the form of true / false or as a numerical comparison result.

[0079] The server combines the results of determining all basic administrative requirements, such as whether septic tanks are insufficient, whether business registration is possible, whether building setback standards are met, and whether setback distances and landscaping areas are satisfied, into a single data set (basic_admin_requirements). This data set is then mapped to the automatic determination values ​​of checklist items during the initial determination data generation stage. Additionally, the data set is inserted as a 'regulatory requirement block' in the LLM input prompt and configured so that the judgment module references it as a basic condition when querying legal grounds. As a result, the server goes beyond simple data collection to perform the automatic determination and standardized delivery of basic administrative requirements based on structural and quantitative criteria.

[0080] A control unit according to one embodiment of the present invention uses the location information to query the building register information in the building administration system (DB), identifies the responses to a plurality of items included in the outdoor business pre-confirmation checklist, and generates primary determination data regarding whether the establishment is eligible for outdoor business or is subject to restriction.

[0081] According to one embodiment of the present invention, the control unit accesses a building administration system (DB) using the location information and searches for a building register record that matches the location coordinates entered through the API or data inquiry interface of the system. The search is performed using at least one method among radius-based inquiry or lot number-based inquiry based on the center point (lat, lon) of the location coordinates, and since the building register DB includes a mapping table between coordinate values, lot numbers, and building management numbers, the control unit obtains accurate register data (e.g., main use, gross floor area, number of floors, structure, attached land information, floor area ratio, etc.) by referring to this.

[0082] The control unit stores data related to the determination of outdoor business operations (e.g., primary use code, regional classification where the building is located, building layout type, site conditions, etc.) extracted from the building register in internal memory, and then combines this with response values ​​included in the outdoor business pre-verification checklist to generate evaluation units for each item. The checklist may be composed of, for example, items such as "inclusion of public open space," "use of open space within the site," "satisfaction of road access conditions," "presence of road occupancy plans," "compliance with the distance from evacuation routes," and "intention to install an outdoor deck or outdoor business area," and each item is displayed in the form of Yes / No (bool), numerical value (number), or multiple selection (category).

[0083] The control unit processes each response of the checklist by normalizing it according to fixed internal rules. For example, yes / no responses are replaced with True / False, numeric values ​​are compared to a threshold to create a flag-type item, and multiple-choice items are converted into one-hot vector data. At this time, the converted data is stored as a primary structured object consisting of "item code, user response value, and ledger data linkage value".

[0084] The control unit generates primary judgment data by combining the normalized items, which refers to data in which evaluation items are reconstructed into a state flag rather than simple selection information. For example, if "main use of the building = neighborhood living facility" and "checklist response = no inclusion of public open space" are simultaneously true, the control unit generates a state variable "basic allowable range satisfied = True"; conversely, if "distance from evacuation passage insufficient" and "included in open space within the site" are detected, it generates a state variable "reason for restriction found = True". Since these state variables involve only simple rule-based operations, they can be easily implemented by an ordinary technician without special training data.

[0085] Once the comparison and normalization process between all checklist responses and ledger attributes is completed, the control unit organizes the state variables into a set and outputs them as a single object called "1st Judgment Data." This 1st Judgment Data serves as the core basis for generating prompts input to the LLM judgment module in subsequent stages. It is configured to store the 'input item ID', 'linked ledger item name', 'comparison result flag', and 'judgment reason text' together during generation so that the basis applied to each state variable can be clearly traced. Therefore, the process of generating the 1st Judgment Data by the control unit is performed by consistent logic based on input data, rather than by arbitrary judgment.

[0086] A control unit according to one embodiment of the present invention identifies the suitability of the main use on the building register (whether it is a neighborhood living facility), whether business registration is possible according to the region or district, the method of connection of the outdoor business area (whether there is direct access on the same floor, whether it is connected within the site), whether evidence proving the right to use the outdoor business area is satisfied, whether facility standards are satisfied, the result of calculating the septic tank capacity (whether it is insufficient compared to the required capacity), whether it is subject to the imposition of the causer pays fee (determination by type of new construction, extension, or change of use), and whether it is subject to mandatory disaster liability insurance based on whether the area is 100㎡ or more, and combines this with the first determination data.

[0087] The above-mentioned control unit matches the primary use code provided from the building register with the Standard Industrial Classification and the table of permitted uses for food service businesses by local government to determine, using a conditional expression, whether the corresponding primary use falls under 'neighborhood living facilities' or a use group where business registration is permitted. For example, it returns a true value if use_code is included in allowed_use_list, and a false value if it is not, thereby recording the suitability.

[0088] In addition, regional and district information is determined using a rule formula by comparing legal Dong codes or district unit plan codes extracted from the building register and location information with the local government regulatory table (local_zone_rules) to determine whether outdoor business operations are permitted in a specific region or district. For example, whether a zone_type is included in the permitted_zones list is recorded as a Boolean value.

[0089] The connection method for outdoor business areas is derived by calculating the movement path between the center point of the outdoor business area and the center point of the indoor business area entrance through CV-based coordinate extraction from the drawing file submitted by the user, and a path estimation algorithm determines whether the two areas are connected by a continuous passageway on the same level. If the floor numbers are not the same, the site layout is referenced to determine whether an external access path exists using the condition path_length > 0.

[0090] Whether the requirements for proof of usage rights are met is determined by identifying the document type of the lease agreement or land use consent form submitted by the user using OCR, and then verifying, based on a comparison table, whether all essential items (party names, addresses, area, and usage consent sections) within the document are extracted, recording the result as true or false.

[0091] Whether facility standards are met is determined by comparing standard values, such as the installation of safety railings and the width of emergency passages, with regulatory tables stored in the storage unit; for example, a conditional expression evaluates whether the passage width `passage_width` extracted from the drawing is greater than or equal to `min_required_width`.

[0092] Septic tank capacity is calculated by setting the required capacity (required_capacity = flow_rate_per_unit × unit_count) as the product of the wastewater generation standard (flow_rate_per_unit) by industry and the number of seats or employees (unit_count), and comparing this with the actual installed capacity (installed_capacity) to record whether required_capacity > installed_capacity as a shortage.

[0093] Whether a project is subject to the developer pays principle is determined by comparing the event codes included in the building register's change history to 'new construction,' 'extension,' or 'change of use,' and if applicable, the necessity for calculating the fee is recorded as true.

[0094] In addition, the area of ​​the outdoor business area is calculated using a polygon area calculation algorithm (area = Σ(x_iy_{i+1} ? x_{i+1}y_i) / 2) from the drawing data submitted by the user, and then the determination of whether the calculated area is 100㎡ or more is made using a threshold comparison method to determine whether there is an obligation to subscribe to disaster liability insurance.

[0095] The above control unit aggregates the result value of each judgment item into a key-value format data object and combines it with the first judgment data, and configures it to be included as a basic administrative requirement block in the LLM input prompt generated thereafter, so that it can be utilized in the next step of the statutory basis judgment process.

[0096] A control unit according to one embodiment of the present invention generates an LLM prompt with the first judgment data as input and queries the LLM-based judgment module, and controls the judgment module to generate one of the judgment results, such as "outdoor business operations allowed," "outdoor business operations restricted," or "additional review required," based on the provisions of the building administration DB and the law and ordinance system (DB), and provides it to the user terminal.

[0097] In order to generate a prompt to be transmitted to an LLM-based judgment module using the primary judgment data as input, the control unit according to one embodiment of the present invention first converts the judgment data into a structured text form including an item name, an input value, a judgment flag, and the basis for occurrence of the corresponding item. In this conversion process, the control unit reconstructs each state variable into a string array of "key:value" pairs and inserts auxiliary text explaining the mutual correlation between the building register data and the checklist response so that the LLM can interpret the context. For example, in addition to simply converted items such as "Main Use = Neighborhood Living Facility," "Includes Public Open Space = False," and "Satisfies Road Access Condition = True," it is configured to also include an explanation of the basis, such as "To determine whether outdoor business operations are permitted, the item is related to Article ○ of the local ordinance."

[0098] The control unit combines the text thus constructed into a single string or a prompt in JSON structure, and the final form of the prompt consists of segments separated into "Judgment Request Type," "Input Data Bundle," "Regulatory Area to Reference," and "Output Format Requirements." For example, the output requirement format is specified to return only one result among "Outdoor Business Permitted / Outdoor Business Restricted / Additional Review Required," and the control unit adds an explicit instruction at the end of the prompt stating, "The result shall be written in a single selection form as one of the three above."

[0099] Upon completion of prompt generation, the control unit executes a query using an internal call API to transmit the prompt to an LLM-based judgment module. At this time, the control unit generates a "Query ID" along with the original data of the prompt to enable subsequent tracking of the judgment results. The judgment module receives the prompt transmitted from the control unit as input and performs a RAG-based processing procedure that references the relevant provisions in the Building Administration DB and the Law and Ordinance System (DB). The judgment module identifies sentences within the prompt that require legal interpretation, converts such sentences into queries to look up relevant statutes and ordinances, and compares the retrieved relevant provisions with the factual context of the prompt to determine whether the permission conditions are met or whether grounds for restriction exist.

[0100] The judgment module compares the retrieved supporting provisions with the item flags in the prompt to select one of "Outdoor business permitted," "Outdoor business restricted," or "Further review required" based on the condition fulfillment rate, conflicts with constraints, and the existence of uncertain items. This judgment is not based on simple keyword matching, but is achieved by verifying, on a sentence-by-sentence basis, the correspondence between specific distance, area, and usage requirements specified in the ordinance and the numerical and flag values ​​transmitted via the prompt. The judgment module generates a structured output object containing the name of the applied supporting provision and the text of the reason for the judgment, along with the final result.

[0101] When the control unit receives a result object returned by the judgment module, it verifies the result type included in the object and then reconstructs and provides the data in a format suitable for the screen size and UI configuration of the user terminal. The provided data consists of contents such as, for example, "Judgment Result = Outdoor business operations permitted," "Basis Article = Local Ordinance Article ○ Paragraph ○," and "Reason for Judgment = Fulfillment of requirements such as road access conditions and non-application of public open space," and the user terminal can immediately display this on the screen.

[0102] A control unit according to one embodiment of the present invention sets importance weights for each item within the LLM prompt so that the primary judgment data is utilized for drawing-based analysis, identification of restricted areas under other laws, and determination of whether ordinance standards are satisfied, configures a judgment path (branch) that the LLM can select according to a combination of judgment items related to adjacent type, usage rights, septic tank, insurance, and charges, and controls the configuration of the judgment path so that it is applied as an essential input condition for the subsequent drawing analysis step and restricted area mapping step.

[0103] A control unit according to one embodiment of the present invention generates a weighting table based on the influence each item occupies in a subsequent stage in order to assign an importance value to each individual item included in the primary judgment data. The weighting table sets items corresponding to, for example, contiguous type, appropriateness of usage rights, suitability of septic tank capacity, whether insurance is mandatory, and whether the causer pays fee is imposed as independent keys, and stores weighting values ​​corresponding to each key as initial setting values ​​or statistical values ​​calculated based on log data of past verification results. The control unit queries the weighting table and, when generating an LLM input prompt, includes an importance tag (e.g., "priority=high", "priority=medium", "priority=low") in front of the item-specific description and supporting data, or adds a lexical emphasis statement in a form that can reflect the judgment weight of each item within the LLM, thereby enabling the primary judgment data to substantially contribute to determining the priority of drawing-based analysis, identification of restricted areas under other laws, and satisfaction of ordinance standards.

[0104] In addition, the control unit utilizes the combination of items from the aforementioned primary judgment data to configure a judgment path (branch) for branching the judgment flow to be selected by the LLM thereafter. More specifically, if the adjoining type is a ground-floor / direct access type, it activates a branch path that prioritizes the review of circulation separation criteria; if evidence proving usage rights is insufficiently identified, it activates a branch path that prioritizes the determination of compliance with ordinance standards. If the septic tank capacity is assessed as insufficient, it is configured to prioritize the analysis of drawing-based infrastructure standards; and if it is determined to be subject to mandatory disaster liability insurance, it adjusts the branch priority so that compliance with insurance obligations carries higher importance than the identification of restricted areas. Furthermore, if it is assessed to be subject to the polluter pays principle, it controls the application of verification of usage and gross floor area, which affect the calculation of the charge, to take precedence over other branches.

[0105] The control unit transmits the judgment path (branch) information configured as described above to the modules called in the drawing analysis stage and the modules called in the mapping stage for restricted areas under other laws, so that the branch information is applied as a mandatory input condition for subsequent stages. At this time, the control unit does not fix the judgment path to a single path, but transmits both a branch activation condition and a branch deactivation condition so that the branch path can be changed if specific conditions are met within the response generated by the LLM. Accordingly, each analysis stage requires the items specified in the judgment path as mandatory inputs, and if the corresponding items are not satisfied, the operation switches to an automatic skip or an auxiliary review module call flow. This control flow provides the effect of improving the consistency and reliability of verification by ensuring that the primary judgment data is not used merely as auxiliary information, but is directly reflected in the judgment order, review scope, and priority review targets of the entire subsequent stage.

[0106] A control unit according to one embodiment of the present invention extracts object information including the location, boundary lines, and entry / exit routes of an outdoor business area included in the drawing file using computer vision (CV) and OCR analysis technology, and integrates the extracted drawing-based information with the building use, gross floor area, structure, and site area stored in the building administration system (DB) into the input prompt of the LLM judgment module.

[0107] A control unit according to one embodiment of the present invention first identifies the format of the file to process the drawing file, determines whether it is in the form of a PDF, an image (JPG, PNG), or a vector-based CAD image, and then applies a preprocessing function suitable for the corresponding format. In the case of an image-based drawing, the control unit is configured to minimize object detection errors in the computer vision analysis stage by performing resolution standardization (e.g., based on 300dpi), tilt correction, and binarization processing.

[0108] Subsequently, the control unit detects basic elements of the drawing, such as lines, polygons, and symbols, using computer vision (CV) analysis technology. For example, to detect boundary lines, it uses a line segment detection algorithm based on Canny edge detection or a line detection algorithm based on Hough transform, and clusters the detected line segments based on coordinates to classify areas forming a closed curve structure as candidate areas for outdoor business premises. The location of the outdoor business premises is converted into (x,y) position pairs in the form of absolute or relative coordinates based on a reference point within the drawing (e.g., the 0,0 coordinate axis or the center coordinate of the building).

[0109] Text information included in the drawing (e.g., "outdoor business area," "deck," "entrance") is extracted using OCR analysis technology, and the control unit automatically matches which object the text describes by calculating the spatial distance (e.g., distance between centroids) or bounding box overlap ratio between the text block and adjacent drawing objects. At this time, the relevance between the object and the text is calculated using cosine similarity-based text embedding comparison or a simple keyword matching method, enabling the reliable identification of the object's meaning even in drawings without explicit labels.

[0110] The control unit forms a polygon by sequentially connecting detected boundary coordinates to calculate the area of ​​the outdoor business area, and derives the area of ​​the polygon using a polygon area calculation formula (e.g., a text implementation of the shoelace formula). To reflect the drawing scale factor during the area calculation, the value is converted to an actual area unit (㎡) by referencing scale labels displayed within the drawing or scale values ​​in the basic CAD information. For entry and exit routes, linear paths connected to entrance symbols or related text blocks within the drawing are extracted, and the length and direction of the path are calculated using coordinate difference values ​​to generate basic data for determining movement feasibility or accident risk factors.

[0111] The control unit integrates the extracted drawing-based information (location coordinates of outdoor business areas, coordinate sets of boundary polygons, area, boundary symbols, types of access routes, etc.) with information on the building's use, gross floor area, structure, and site area retrieved from the building administration system (DB). During the integration process, in order to combine all data into a single prompt structure, the control unit normalizes the drawing analysis information into the "drawing_info" field and the building administration DB information into the "building_registry_info" field, respectively, and reconstructs them in the form of arrays or key-value pairs.

[0112] Within the prompt, each data point is represented in a "ItemName=Value" structure or a JSON-like "{key:value}" structure so that the LLM can reference it, and the matching relationship between drawing information and building administration database information is explicitly linked by adding "relationship description text." For example, judgment criteria are included as text, such as "Outdoor business area = 18.4㎡, comparison with ordinance-approved area conditions required."

[0113] This integrated prompt is used as an essential input value when the LLM judgment module performs judgments based on laws and ordinances, and the control unit is configured by applying the sequence of "1) factual circumstances, 2) drawing-based information, 3) basic building information, 4) content of the judgment request" to maintain a logical flow between items within the prompt. This provides sufficient information to enable the LLM judgment module to accurately interpret whether outdoor business operations are permitted based on the relevant provisions of the Building Administration DB and the Law and Ordinance System (DB).

[0114] A control unit according to one embodiment of the present invention performs drawing-based area calculation, determines whether circulation paths are separated, determines whether there is an obstruction to an emergency passage, determines whether restricted areas under other laws regarding public open spaces, light areas, evacuation passages, road occupancy areas, and mandatory parking areas are included, and identifies conditions requiring confirmation by departmental personnel. Based on the results identified by the judgment module, it controls the generation and transmission to the user terminal of warning data that includes the display of the possible outdoor business area and scope on the drawing when permitted, and the name of the violating statute, article number, grounds, and contact information of the relevant department when restricted.

[0115] A control unit according to an embodiment of the present invention calculates the area of ​​an outdoor business premises using object information extracted from a drawing file. First, the control unit converts the outline coordinates detected through computer vision analysis into a polygonal coordinate sequence and calculates the actual area on the drawing by continuously connecting the boundary lines between each point included in the coordinate sequence. At this time, the control unit applies a conversion factor (scale_factor) for actual area conversion based on the drawing file metadata or scale information, and the conversion factor is calculated using the ratio of the reference line length indicated on the drawing to the pixel length on the image. The calculated area is stored as a structured field so that it can be used for comparison with outdoor business permit standards (e.g., stricter application of fire and sanitation standards when the area exceeds a certain threshold).

[0116] The control unit then determines whether the entry and exit routes are separated by dividing the entrance and exit objects and passageway objects indicated in the drawing to extract their respective centerlines, and compares the minimum distance values ​​between the paths to determine whether the entry and exit routes are separated or intersect and overlap. In addition, to determine whether there is an obstruction in the emergency passage, the path width of the emergency passage object is sampled at regular intervals to check if there is a section smaller than the minimum allowable width standard, and if obstacle objects such as furniture, partitions, or boundary markers overlap with the emergency passage path, they are identified as obstruction elements.

[0117] Whether a restricted area under other laws, such as public open space, restricted area, evacuation passage, road occupancy area, and mandatory parking area, is included is determined by the control unit querying pre-stored GIS-based area boundary data or spatial coordinate data from the building administration system (DB) to determine the inclusion relationship between the coordinates of the outdoor business area on the drawing and the boundary coordinates of the restricted area. The control unit manages the boundary surface of the restricted area as polygon or multipolygon data and determines whether the coordinate sequence of the outdoor business area is included within the said boundary surface using a point-in-polygon method. The determination result is stored to be linked to reference metadata, such as the name of the relevant statute and applicable standard conditions (e.g., separation distance, occupancy restriction range), along with a simple inclusion / exclusion status.

[0118] The identification of conditions requiring confirmation of personnel by department is performed by the control unit by comparing specific items among the results of the drawing analysis (e.g., narrow emergency passage, possibility of occupying public open space, need to review special exceptions for the method of connecting outdoor business areas, etc.) with a predefined list of conditions subject to review. The control unit stores a mapping table with the responsible department corresponding to each condition as the key, and when a condition is identified, it structures information including the name of the responsible department, contact information of the person in charge, and the basis of administrative procedures by referencing the corresponding key.

[0119] Once the drawing-based analysis and restricted area identification are completed as described above, the control unit combines this with the results generated by the judgment module to construct output data for the end user. If it is determined that outdoor business operations are permitted, the control unit generates a rendering image in which the permitted area is indicated by a translucent color or a boundary overlay to visually display the calculated possible outdoor business area and the allowable range on the drawing image. This allows the user to intuitively verify the actual installation locations.

[0120] Conversely, if it is determined that a restriction is necessary, the control unit generates warning data containing the name of the violating statute related to the reason for the restriction, the article number, standard phrases (e.g., insufficient separation distance, prohibition of occupying public open space, etc.), a summary of the grounds, and the contact information of the responsible department. This warning data is generated in a JSON or key-value structure and transmitted to the user terminal; on the user screen, links to specific articles, emphasis on the grounds, and information regarding whether action is required are displayed automatically. Through such control operations, drawing-based analysis and determination of administrative requirements can be delivered to the user immediately, and since the grounds for judgment are presented in a clear format, the reliability and predictability of the administrative review are enhanced.

[0121] A control unit according to one embodiment of the present invention determines, based on pre-set objective criteria, direct accessibility between an entrance and an outdoor business area on a drawing, whether there is movement between floors and whether there is contiguousness within the site, and whether there is intervention of other use spaces in common spaces, corridors, and parking lots; calculates the overlap ratio with restricted areas under other laws and the center coordinate error range to quantify the possibility of inclusion in a restricted area; classifies items requiring review by department corresponding to each restriction condition and generates a structured restriction condition dataset that can be transmitted together with the judgment results.

[0122] A control unit according to an embodiment of the present invention first compares the center coordinates of an entrance object extracted by computer vision analysis with the boundary coordinates of an outdoor business area to determine direct accessibility between an entrance and an outdoor business area on a drawing. To evaluate the straight-line distance and the continuity of the movement path between the two objects, the control unit models the boundary line on the image in the form of a graph and searches for the shortest path from the center point of the entrance to the boundary line of the outdoor business area. The control unit determines that direct access is possible if this path is less than or equal to a preset allowable travel distance standard and there are no corridor, stair, or obstacle objects on the path, and classifies it as indirect access or inaccessible if at least one of the travel distance or obstacles exceeds the standard.

[0123] The determination of whether there is inter-floor movement is performed by combining the floor number metadata of a drawing file or a building register with the existence of inter-floor movement elements (stairs, elevators, etc.) indicated on the drawing. The control unit first checks whether an entrance and an outdoor business area are located on the same floor plan, and if the objects exist on different floors, it determines whether inter-floor connection is possible based on the existence of a dedicated circulation path between floors, the total length of the circulation path when using a dedicated circulation path, and the independence of the circulation path (e.g., whether it passes through other use spaces such as a planting space, parking lot, or dedicated corridor).

[0124] The determination of whether objects are adjacent within the site is performed by comparing the site boundary polygon data stored in the building administration system (DB) with the outdoor business area polygon on the drawing, and the control unit uses whether the two polygons are in contact or whether they share a boundary surface as a criterion for determining the connection between objects. To this end, the control unit calculates the minimum distance and whether there is an intersection between each segment of the two polygons, and identifies them as adjacent if the minimum distance is 0 or less than or equal to the allowable error range epsilon.

[0125] The determination of whether other use spaces interfere with common areas, corridors, and parking lots is performed by the control unit determining whether the boundary lines of objects classified as "common," "corridors," or "parking lots" among the space objects extracted from the drawings overlap with the centerlines of the access routes. If a route passes through such common areas, the control unit checks the use, area, and risk factors (e.g., the presence of vehicle traffic routes) of the relevant space and determines whether it conflicts with the conditions requiring independence as a commercial space under administrative standards.

[0126] The overlap ratio with restricted areas under other laws is calculated by the control unit calculating the intersection area between the restricted area polygon and the outdoor business area polygon, and then calculating the ratio to the total outdoor business area. The above ratio is calculated using the following text formula.

[0127] overlap_ratio = intersection_area / total_area

[0128] The center coordinate error is calculated using the Euclidean distance between the center point of the restricted area and the center point of the outdoor business area, and if the said distance is less than or equal to the preset error range threshold_distance, it is determined that there is a possibility that the two areas substantially overlap.

[0129] distance = sqrt( (x1 - x2)^2 + (y1 - y2)^2 )

[0130] The control unit generates a restriction score that quantifies the possibility of inclusion in a restricted area by combining the overlap_ratio and distance, and this score is used by the judgment module in a subsequent step to evaluate the strength of administrative restrictions.

[0131] The control unit classifies items requiring departmental review corresponding to each restriction condition by referring to a condition-based mapping table. For example, the system is configured to connect the Urban Planning Department when public open space overlap is detected, the Fire and Safety Department when evacuation route overlap is detected, and the Traffic or Construction Department when mandatory parking area overlap is detected. The control unit generates a structured restriction dataset (restriction_dataset) from the above condition-based mapping results in a key-value format (e.g., "condition_type: Evacuation Route", "department: Fire and Safety Department", "contact: Phone Number"), and serializes the dataset into a form suitable for a communication data format (such as JSON) so that it can be transmitted to a user terminal along with the judgment results.

[0132] The constraint dataset configured in this manner is designed to allow users to verify the reasons for restrictions and immediately identify items requiring consultation between administrative departments, thereby contributing to enhancing the explainability of the judgment module's results and the predictability of licensing.

[0133] A control unit according to one embodiment of the present invention identifies whether the location of the building or outdoor business premises is included in any one of a public open space, an open space within the site, a light zone, an evacuation passage, a road occupancy zone, or a mandatory parking area by matching the location information, floor plan, and layout data of the building with a previously stored database of restriction conditions under other laws.

[0134] When a control unit according to one embodiment of the present invention receives location information, floor plan, and layout data of the building as input, it first normalizes the location information data into a coordinate system-based GIS (Geographic Information System) standard format. If the location information is in the form of address information, it calls a geocoding engine to convert it into latitude and longitude coordinates, and if it has already been provided as a coordinate value, it maps the corresponding coordinates directly to an internal coordinate object.

[0135] Normalized coordinate information is utilized for spatial matching with a database of restriction conditions under other laws, and the said restriction database stores areas corresponding to public open space, open space within the site, boundary zones, evacuation passages, road occupancy zones, and mandatory parking areas, respectively, as spatial objects in the form of polygons or line strings. The control unit calls the 'point-in-polygon' function provided by the GIS library to calculate whether the reference point of the building (e.g., the building center coordinates or the representative point of the boundary polygon) is included in any of the said polygon areas. The inclusion status is returned as a Boolean value, and the type of included area is determined through the attribute value assigned to the corresponding polygon object.

[0136] In addition, the control unit converts objects within the drawings into a vector-based coordinate system to analyze the geometric structure of the floor plan and layout plan. Building boundary lines identified in the layout plan are restored in the form of polygons, and the outer space that can be used as an outdoor business area is separated based on a specific color, line thickness, and symbol through a CV-based extraction algorithm. The control unit converts the coordinates of each vertex of the restored building boundary lines to the same standard as the GIS coordinate system to further determine whether the entire boundary overlaps with the restricted area polygon. The overlap is determined by a 'polygon-polygon intersection' function, and if they intersect, the intersect area is also calculated to quantitatively determine the degree of collision with the restricted area.

[0137] Subsequently, the control unit integrates the result of the above spatial operation to determine whether it is "included in a restricted area." For example, if the point-in-polygon result is true or the intersect_area value in the polygon intersection operation is greater than 0, it is determined that the building or outdoor business premises are included in a specific restricted area. The control unit identifies the attributes of the included restricted area (public open space, open space within the site, light zone, evacuation passage, road occupancy area, or mandatory parking area), and together looks up the inspection code and the relevant statute / article number corresponding to the attribute to generate a structured determination result to be transmitted to a deep verification or warning data generation module in a subsequent stage.

[0138] A control unit according to one embodiment of the present invention generates warning data containing the name of the relevant statute, article number, name of the responsible department, and contact information for each item where a violation risk exists when included in the restricted area, transmits the warning data to the electronic signature module of each relevant department to collect signatures, and controls the issuance of a license verification result report to the user terminal once signatures from all relevant departments are completed.

[0139] When a control unit according to an embodiment of the present invention determines that a location is included in the restricted area, it first queries the statute name, article number, name of the responsible department, and contact information pre-stored in a database corresponding to the type of restricted area. At this time, the query is configured to match using the restricted area attribute value as a key; for example, if the restricted area is identified as "public open space," it returns a corresponding record containing "Article ○○ Paragraph ○ of the Building Act," "Architecture Department," "main number," etc. The control unit structures the queried items to create a warning data object, and the warning data consists of data in JSON or XML format including a restricted item identifier, the name of the relevant statute, the article number, the name of the responsible department, the contact information of the responsible department, and an electronic signature request code.

[0140] The generated warning data is transmitted to the digital signature modules equipped in each relevant department. The control unit queries the API endpoints of the digital signature modules based on the unique department code distinguishing each department and transmits the warning data to individual modules using HTTPS-based communication. Each digital signature module parses the digital signature request code within the warning data to identify the items to be signed, generates signature data using the digital signature key used by the person in charge of the relevant department, and sends it back to the control unit. To verify whether the received signature data was generated successfully without tampering, the control unit executes a signature data verification algorithm (e.g., a public key-based verification routine) and verifies integrity by comparing whether the signed hash value matches the hash of the original data included in the warning data.

[0141] The control unit calculates the required number of signatures (N_total) based on the entire list of related departments, updates the number of received signatures (N_received) whenever a signature is returned, and determines that signature collection is complete when N_received equals N_total. If signatures from all departments have not arrived, a retransmission routine for non-responding departments is executed at regular intervals, and the number of retransmissions and the timeout value are controlled according to the configuration data in the storage unit.

[0142] When signatures from all relevant departments are successfully collected, the control unit generates a licensing verification result report to provide the final verification results to the user terminal. This report consists of structured data including the restricted area determination results, the names and article numbers of the relevant laws for each warning item, whether signature collection is complete, a list of signing departments, and a verification timestamp; it is converted into HTML, PDF, or a dedicated viewer format so that it can be displayed directly on the user terminal screen. During conversion, each item in the report is arranged according to the field mapping rules recorded in the storage unit, and if the location or scope of the restricted area exists as drawing-based data, it is provided in a form where visual markers (colors, labels) are overlaid on the corresponding drawing image so that the user can verify it immediately.

[0143] A control unit according to one embodiment of the present invention calculates the statistical similarity and pattern matching rate for each object by comparing the location coordinates and attribute data of an object extracted from the drawing file with first historical data corresponding to the same industry, similar buildings, and similar districts among the log data of past permit verification results accumulated in the storage unit.

[0144] A control unit according to an embodiment of the present invention receives location coordinates and attribute data of an object extracted from the drawing file as input, and first establishes a reference coordinate system to normalize the coordinates of each object. The reference coordinate system is generated by setting a global reference point within the drawing file (e.g., a 'site reference point' on the layout plan) as the origin (0,0) and converting all coordinates to the same unit (meter unit) using the scale information of the drawing. The converted coordinate values ​​are stored in an array form so that quantitative comparison with historical data is possible thereafter.

[0145] The control unit utilizes the normalized object coordinates and attribute data to retrieve first historical data corresponding to the same industry, similar buildings, and similar districts from the log data of past permit verification results accumulated in the storage unit. The first historical data is searched using keys such as industry codes, building type codes, and district unit planning area codes, and is configured to include an object list recorded in past actual verification procedures, the location coordinates of each object, object types (e.g., tables, entrances / exits, boundary lines), area values, movement path data, and verification result tags (allowed / restricted). Since this first historical data is based on actual measurements accumulated during the administrative processing, it ensures statistical significance and can be used as a standard for comparison.

[0146] The control unit performs a matching process to correspond the two datasets one-to-one in order to calculate the statistical similarity between the position coordinates of the current drawing-based objects and the position coordinates of the first history data. The matching process is performed by using the object type as the primary criterion and matching the closest object between objects of the same type using the distance between coordinates. The distance value is calculated using the Euclidean distance method sqrt((x1 - x2)^2 + (y1 - y2)^2), and if the distance value is less than or equal to a set tolerance, it is determined that a match has been made.

[0147] The control unit calculates the position similarity (score_pos) based on distance values ​​for matched object pairs. For example, it calculates the score by linearly transforming it so that score_pos becomes closer to 1 as the distance value approaches 0, and score_pos becomes closer to 0 as the distance value approaches the maximum allowable value. In addition, it calculates the difference ratio for object-specific attribute data (area value, boundary line length, entry / exit direction, object placement angle, etc.) and calculates the attribute similarity (score_attr) based on this difference ratio.

[0148] The control unit generates the final statistical similarity (similarity_total) through a combination of score_pos and score_attr calculated for each object. Similarity_total is calculated by combining two similarity items according to a fixed weight (weight_pos, weight_attr), for example, in the form of similarity_total = (weight_pos × score_pos) + (weight_attr × score_attr). These weights are values ​​automatically calculated based on log data of past verification results recorded in the storage unit, and since dynamically adjusted values ​​are used depending on which item, such as location error or area error, has a stronger influence on the verification results in actual administrative experience, the similarity calculation process is not arbitrary.

[0149] In addition to the aforementioned similarity, the control unit independently calculates the pattern match rate. To compare the overall layout structure of objects, the pattern match rate is calculated by converting the overall spatial arrangement pattern on the drawing—such as the grouping results by object type, the relative positional relationship between entrances / exits and main circulation paths, and the shape of outdoor business area boundaries—into a graph structure for comparison. The control unit constructs an adjacency matrix between the current drawing-based graph and the first history data-based graph, and then calculates the pattern match rate by comparing the degree of agreement between the two matrices. This value is normalized to a range of 0 to 1, and a value closer to 1 indicates a layout pattern that is nearly identical to the past history.

[0150] A control unit according to one embodiment of the present invention continues the verification procedure only when the similarity and pattern matching rates each exceed a preset threshold, and controls the building to be classified as a subject for in-depth review when the similarity or pattern matching rate is below the preset threshold.

[0151] A control unit according to an embodiment of the present invention receives location coordinates and attribute data of an object extracted from the drawing file as input, and first establishes a reference coordinate system to normalize the coordinates of each object. The reference coordinate system is generated by setting a global reference point within the drawing file (e.g., a 'site reference point' on the layout plan) as the origin (0,0) and converting all coordinates to the same unit (meter unit) using the scale information of the drawing. The converted coordinate values ​​are stored in an array form so that quantitative comparison with historical data is possible thereafter.

[0152] The control unit utilizes the normalized object coordinates and attribute data to retrieve first historical data corresponding to the same industry, similar buildings, and similar districts from the log data of past permit verification results accumulated in the storage unit. The first historical data is searched using keys such as industry codes, building type codes, and district unit planning area codes, and is configured to include an object list recorded in past actual verification procedures, the location coordinates of each object, object types (e.g., tables, entrances / exits, boundary lines), area values, movement path data, and verification result tags (allowed / restricted). Since this first historical data is based on actual measurements accumulated during the administrative processing, it ensures statistical significance and can be used as a standard for comparison.

[0153] The control unit performs a matching process to correspond the two datasets one-to-one in order to calculate the statistical similarity between the position coordinates of the current drawing-based objects and the position coordinates of the first history data. The matching process is performed by using the object type as the primary criterion and matching the closest object between objects of the same type using the distance between coordinates. The distance value is calculated using the Euclidean distance method sqrt((x1 - x2)^2 + (y1 - y2)^2), and if the distance value is less than or equal to a set tolerance, it is determined that a match has been made.

[0154] The control unit calculates positional similarity (score_pos) based on distance values ​​for matched object pairs. For example, it calculates score_pos by applying a linear transformation such that score_pos becomes closer to 1 as the distance value approaches 0, and score_pos becomes closer to 0 as the distance value approaches the maximum allowable value. In addition, it calculates a difference ratio for object-specific attribute data (area value, boundary line length, entry / exit direction, placement angle of objects, etc.) and calculates attribute similarity (score_attr) based on this difference ratio.

[0155] The control unit generates the final statistical similarity (similarity_total) through a combination of score_pos and score_attr calculated for each object. Similarity_total is calculated by combining two similarity items according to a fixed weight (weight_pos, weight_attr), for example, in the form of similarity_total = (weight_pos × score_pos) + (weight_attr × score_attr). These weights are values ​​automatically calculated based on log data of past verification results recorded in the storage unit, and since dynamically adjusted values ​​are used depending on which item, such as location error or area error, has a stronger influence on the verification results in actual administrative experience, the similarity calculation process is not arbitrary.

[0156] In addition to the aforementioned similarity, the control unit independently calculates the pattern match rate. To compare the overall layout structure of objects, the pattern match rate is calculated by converting the overall spatial arrangement pattern on the drawing—such as the grouping results by object type, the relative positional relationship between entrances / exits and main circulation paths, and the shape of outdoor business area boundaries—into a graph structure for comparison. The control unit constructs an adjacency matrix between the current drawing-based graph and the first history data-based graph, and then calculates the pattern match rate by comparing the degree of agreement between the two matrices. This value is normalized to a range of 0 to 1, and a value closer to 1 indicates a layout pattern that is nearly identical to the past history.

[0157] A control unit according to one embodiment of the present invention, when classified as a subject for in-depth review, calls a self-correction module provided in the control unit to extract cases among past verification result log data where the false positive rate or error rate is greater than or equal to a threshold value as training data, performs retraining on correction coefficients and weights using the training data, updates and stores the retrained correction coefficients and weights in the storage unit, and controls the system to improve the reliability of verification results by applying the updated correction coefficients and weights to future verification procedures for the same industry or similar buildings.

[0158] When the building is classified as a subject for in-depth review according to one embodiment of the present invention, the control unit calls a self-correction module included within the control unit to initiate a procedure for relearning correction coefficients and weights. The control unit first retrieves the final judgment value, the actual administrative processing result value, and the comparison reference value recorded for each verification case from the past verification result log data accumulated in the storage unit, and constructs a training dataset by filtering only cases where the false positive rate or error rate is recorded above a reference value from these data. This filtering is performed based on whether the value calculated in the manner of error_rate = |system_decision - actual_decision|, which is calculated by calculating the difference between the correct label value (actual administrative judgment) and the system judgment value included in the log data, is above the stored reference value.

[0159] The control unit transmits the extracted training dataset to the self-correction module, which performs an update operation for the corrective factor by correlating the input characteristics stored in each data item (drawing-based object attributes, location coordinates, industry code, similarity index, pattern matching rate, etc.) with the corresponding item's error_rate. This operation is carried out using a linear regression-based or gradient-based update method to adjust the corrective factor so that it directly reduces the error, and the corrective factor and weight value are updated in the direction of decreasing the error_rate at each epoch. During this process, the self-correction module analyzes the contribution of each characteristic and reconstructs the model by assigning higher weights to items that have a greater influence on reducing the error_rate, while gradually decreasing the weights of items with low contribution.

[0160] When retraining is completed, the self-correction module transmits the updated correction coefficients and weights to the control unit, and the control unit immediately records them in the storage unit so that the newly learned correction coefficients and weights can be applied in subsequent verification procedures for the same industry, similar buildings, and similar districts. When storing, the difference from the existing value, the size of the dataset used for training, and the number of training iterations are also logged so that they serve as a basis for determining reliability during subsequent additional verification and correction processes.

[0161] The control unit applies updated correction coefficients and weights to adjust the judgment logic in a way that reduces the false positive or error rate compared to the past when performing verification under identical or similar conditions in the future. In other words, input characteristics that tended to cause errors in the past are processed in a manner where correction coefficients are weighted, preventing the repetition of errors during the calculation of the final judgment value; this is designed so that such individual characteristic-based adjustments result in a gradual improvement in overall judgment reliability. Accordingly, since the retraining procedure of this embodiment is based on quantitative error data directly collected from past verification cases, the data structure itself demonstrates that the updating of correction coefficients is performed in a data-based manner reflecting actual error occurrence patterns, rather than through arbitrary value setting.

[0162] A control unit according to one embodiment of the present invention additionally calls an automatic basis provision tracking module for a judgment result generated by the LLM-based judgment module, and the automatic basis provision tracking module converts item-unit text blocks of individual laws, ordinances, notices, and guidelines stored in the building administration system (DB) and the law and ordinance system (DB) into a basis relevance score in a vector embedding space, and calculates the semantic similarity score and logical entailment score, respectively, between the expression within the sentence generated by the judgment module (e.g., "unable to secure access route," "occupancy of public open space") and the text blocks.

[0163] A control unit according to one embodiment of the present invention immediately calls an automatic basis provision tracking module when a judgment result generated by the LLM-based judgment module is output. The automatic basis provision tracking module first calls item-unit text blocks of individual laws, ordinances, notices, and guidelines stored in the building administration system (DB), the law DB, and the ordinance DB. The control unit divides each text block into sentences or items and generates text block vectors by providing each divided text block as input to a pre-constructed embedding model (e.g., a sentence vector generation model). The embedding model is trained based on experimental data such that items with the same meaning are placed in adjacent positions within a vector space, and each item is converted into a vector representation of the same number of dimensions.

[0164] The control unit also generates a judgment sentence vector by inputting the judgment sentences (e.g., "Unable to secure access route," "Occupancy of public space") generated by the judgment module into the same embedding model. Subsequently, the basis provision automatic tracking module calculates the semantic similarity (similarity_score) by calculating the cosine similarity between the judgment sentence vector and each text block vector. The cosine similarity is calculated using an operation of the form dot(a,b) / (sqrt(dot(a,a)) * sqrt(dot(b,b))), and the closer the similarity value is to 1, the more semantically close the judgment sentence and the text block are.

[0165] Furthermore, the above-mentioned automatic tracking module for the basis clause calls a text-to-text entailment judgment model to identify the logical attribution relationship between the judgment sentence and each text block, and outputs an entailment_score indicating whether the judgment sentence is directly grounded by a specific text block. The entailment_score returns a probability value indicating which of the three categories the judgment sentence belongs to by the item is 'directly grounded', 'related but not direct ground', or 'contradictory', and is expressed as a value closer to 1 as the likelihood of logical attribution increases.

[0166] The automatic basis provision tracking module calculates a relevance score for individual text blocks by combining these similarity_score and entailment_score. The relevance score is calculated as a weighted combination of the two values, and is configured so that items that contributed significantly to actual administrative decisions in verified data logs are automatically assigned higher weights. Since the relevance score is an indicator that reflects both semantic similarity and logical attribution between the judgment sentence and the text block, the control unit filters only text blocks with a relevance score above a threshold value and ultimately selects them as the basis for the corresponding judgment result.

[0167] Since the above series of procedures is configured to search text blocks within the DB based on sentence meaning rather than simple keywords, a high relevance_score is generated when the administrative intent, regulatory purpose, and contextual information are identical, even if the terms of the judgment sentence and the statutory item do not completely match. As a result, the data processing process itself demonstrates that the configuration of this embodiment automatically extracts only the provisions of laws and ordinances selected through actual meaning-based matching as the basis for judgment, rather than a method of arbitrarily selecting provisions.

[0168] A control unit according to one embodiment of the present invention automatically maps statute items in which the final basis score (final_basis_score), obtained by weighting the relevance_score, similarity_score, and entailment_score, is greater than or equal to the preset reference score to judgment results of 'outdoor business permitted', 'outdoor business restricted', and 'additional review required', respectively, and for each mapped statute item, structures the item number, the conditional statement structure of a specific clause (e.g., based on provisos, main text, or appendices), and the scope of application of ordinances that differ by region in a key-value format.

[0169] A control unit according to an embodiment of the present invention calculates a final basis score (final_basis_score) by weighting each score together with a pre-set weight when the relevance_score, similarity_score, and entailment_score are calculated. The final_basis_score is calculated in the form of relevance_score × w₁ + similarity_score × w₂ + entailment_score × w₃, and each weight (w₁, w₂, w₃) is set based on real data so that items that contributed more to the actual judgment in past administrative verification logs are automatically corrected to higher values. The control unit treats the relevant legal item as a valid basis only when the final_basis_score is equal to or greater than a pre-set threshold score, and automatically maps the valid basis item to correspond to one of the judgment results output by the judgment module: "Outdoor business permitted," "Outdoor business restricted," or "Further review required." During the mapping process, the control unit tokenizes key expressions within the judgment sentences generated by the judgment module (e.g., "satisfies permissible range," "includes restricted area," "requires additional review under the law") to break them down into major semantic units, compares the matching priority between each semantic unit and statutory items whose final basis score is equal to or greater than the threshold score, and connects the judgment result with the statutory item that exhibits the highest correlation.

[0170] Subsequently, the control unit separates the mapped statute items by data field—specifically the item number, the structure of the conditional statement of a specific clause (e.g., whether it corresponds to a proviso, the main text, or a detailed item under the attached criteria), and the range setting of ordinances that apply differently by region (e.g., whether the clause applies only to a specific autonomous district)—and structures them in a key-value format. The structuring process is performed by generating fixed keys such as "article_number", "clause_type", "exception_flag", "annex_reference", and "local_rule_scope", and mapping actual values ​​extracted from the item (e.g., "Article 12, Paragraph 3, main text", "Attached Table 2, Criterion 1", "Applicable to Seoul Jung-gu Ordinance Criteria") to corresponding values. The control unit automatically classifies the sentence structure of each item into conditional statements, proviso statements, and main text, and identifies the scope of ordinance application by comparing whether the administrative district code within the database matches the corresponding district name within the item.

[0171] As a result of such processing, multiple legal items corresponding to a single judgment result are generated as structured data in the form of JSON with a key-value structure or a table, which can be reused in a user terminal or a subsequent verification module. According to the present embodiment, the judgment result is mapped based on the semantic basis of the legal items, and the clause number, conditional statement structure, and regional ordinance application scope of the mapped legal items are all converted into a machine-readable structure, thereby eliminating arbitrariness or ambiguity in the judgment process, and it is clearly revealed through the data processing flow alone that a person skilled in the art can implement the same structure as is.

[0172] A control unit according to one embodiment of the present invention generates a basis_trace_dataset combined with the result of the judgment module, and if the basis_trace_dataset exceeds a threshold_deviation when compared with the past judgment result of the first history data for the same industry, similar building, or similar district, controls the system to correct at least one of the prompt configuration method, RAG search range, and DB indexing weight of the LLM-based judgment module in conjunction with the self-correction module, and transmits the judgment result regenerated after correction and the basis_trace_dataset to the user terminal to output the difference (diff) between the new judgment result and the judgment result before correction and the history of changes in the basis clause in the form of an automatic comparison table.

[0173] A control unit according to an embodiment of the present invention generates a basis trace dataset (basis_trace_dataset) by combining a judgment result generated by the judgment module and basis trace metadata including a traced basis provision and a semantic similarity score corresponding to the judgment result. The basis trace dataset is configured to include a core judgment sentence output by the judgment module, a major token unit representation of the judgment sentence, a matched legal provision identifier, a basis score for each provision, a RAG search range at the time of judgment, and a prompt template ID used as a single structured data bundle. This configuration can be implemented as key-value data in JSON format, and a person skilled in the art can easily generate it by structuring the string output by the judgment module and the basis provision matching result as is.

[0174] The control unit calculates a threshold deviation to determine whether a difference occurs between judgment results when the aforementioned evidence tracking dataset is compared with first historical data performed in the past for the same industry, similar buildings, or similar districts. The threshold deviation is derived by comparing at least one of the following: the semantic difference between past judgment sentences and new judgment sentences (semantic_shift), the consistency between matched legal provisions (consistency_rate), and the change in the evidence score (score_delta). For example, semantic_shift can be calculated as the distance between text embedding vectors in the form of cosine_distance, consistency_rate can be calculated as the ratio of the intersection between the past and new sets of provisions, and score_delta can be measured as the absolute value (abs()) of the difference in evidence scores. If these values ​​are greater than the stored threshold value, the control unit determines that the threshold deviation has been exceeded, and is configured to consider the repeated occurrence of such deviation not as a simple case deviation, but as a structural problem in the input prompt configuration of the judgment module or the setting of the RAG search range.

[0175] When the above-mentioned standard deviation is exceeded, the control unit calls the self-correction module to control the correction of at least one of the operating elements of the LLM-based judgment module, such as the prompt configuration method, RAG search range, and DB indexing weight. Correction of the prompt configuration method can be performed by using a prompt structure that demonstrated high judgment accuracy in past first-history data (e.g., order of clause citation, order of listing building attributes, method of expressing checklist items) as a template, and by rearranging the sentence placement order within the new input prompt or changing the weighted placement of specific attributes. Correction of the RAG search range is performed by relocating the relevant set of clauses to a priority search group or expanding the search range when instances are found where specific local government ordinances or parts of specific laws have been repeatedly omitted from the judgment. Correction of DB indexing weight can be implemented by raising the indexing priority for clause groups that repeatedly demonstrated high matching accuracy in the past history, and slightly lowering the weight for items with low accuracy.

[0176] After performing the correction process, the control unit calls the LLM judgment module again for the same input data to regenerate the corrected judgment result and the corrected evidence tracking dataset, and compares them with the previous judgment result and the previous evidence tracking dataset to organize the differences (diff) in the form of an automatic comparison table. The automatic comparison table is configured in a table form in which "whether the prompt configuration has changed," "whether the search range has changed," "whether the matched legal provision has changed," "amount of change in the evidence score," and "whether the final judgment sentence has changed" are arranged in rows for each item, and the data before and after correction is arranged in columns. The control unit transmits the comparison table to a user terminal so that the correction effect can be visually verified. This processing method demonstrates that the present invention is a real data-based structure that tracks and compares changes in evidence rather than simply outputting results, so a person skilled in the art can implement the flow exactly as is, and it is clearly revealed that it is not an arbitrarily created procedure. Explanation of the symbols

[0180] 100: User terminal 110: User Input Section 120: Display section 200: Server 210: Communications Department 220: Storage section 230: Control unit< / checklist> < / location>

Claims

Claim 1 In an AI-based automatic determination of eligible outdoor business targets and licensing verification system, a user terminal equipped with a display unit and a user input unit; and a server connected to the user terminal via a wired or wireless network; wherein the server, upon receiving building location information, drawing files, building register, business operator information, and outdoor business pre-confirmation checklist data entered by the user, converts the received data into structured input data through a prompt generation module that converts the received data into an LLM input prompt, and is configured to determine whether the relevant building is eligible for outdoor business by querying basis information linked to the building administration system (DB) and the law and ordinance system (DB) in a RAG manner using an LLM-based judgment module, and the server is configured to automatically determine whether there is an obligation to subscribe to disaster liability insurance based on the location, number of floors, and area information of the business establishment, and to calculate the normalization results of the gross floor area and usage standards required for calculating the causer pays fee and whether the use is permissible for food service businesses based on building register data, and the server is configured to provide basic administrative requirements including septic tank capacity calculation standards (whether there is a shortage compared to the required capacity), the scope of business reporting per industry code, setback standards for the site and building line, and whether there are restrictions on separation distance and landscaping area according to local government ordinances. An AI-based automatic determination of outdoor business eligibility targets and licensing verification system characterized by collecting and normalizing judgment data and adding it to the structured input data, and configuring the basic administrative requirement judgment data to be utilized as a component of the primary judgment data and LLM input prompt generated in a subsequent stage. Claim 2 In paragraph 1, the server comprises a communication unit; a storage unit; and a control unit; wherein the control unit uses the location information to query the building register information in the building administration system (DB), identifies responses to multiple items included in the outdoor business pre-confirmation checklist to generate primary determination data regarding whether the establishment is eligible for outdoor business or subject to restrictions, identifies the suitability of the main use in the building register (whether it is a neighborhood living facility), the possibility of business registration according to the region / district, the method of contiguous access to the outdoor business area (whether there is direct access on the same floor, whether it is contiguous within the site), whether evidence proving the right to use the outdoor business area is satisfied, whether facility standards are satisfied, the result of septic tank capacity calculation (whether it is insufficient compared to the required capacity), whether it is subject to the polluter pays principle (determination by type of new construction, expansion, or change of use), and whether it is subject to mandatory disaster liability insurance based on whether the area is 100㎡ or more, combines this with the primary determination data, generates an LLM prompt with the primary determination data as input to query the LLM-based judgment module, and the judgment module [reviews] the building administration DB and laws / ordinances An AI-based automatic determination of outdoor business eligibility targets and licensing verification system characterized by controlling the generation of one of the judgment results—"outdoor business possible," "outdoor business restricted," or "additional review required"—based on the provisions of the system (DB) and providing it to the user terminal, wherein the control unit sets importance weights for each item within the LLM prompt so that the primary judgment data is utilized for drawing-based analysis, identification of restricted areas under other laws, and determination of whether ordinance standards are satisfied, configures a judgment path (branch) that the LLM can select according to a combination of judgment items related to adjacent type, usage rights, septic tank, insurance, and charges, and controls the configuration of the judgment path so that it is applied as an essential input condition for the subsequent drawing analysis stage and restricted area mapping stage. Claim 3 In paragraph 2, the control unit extracts object information including the location, boundary lines, and entry / exit routes of the outdoor business area included in the drawing file using computer vision (CV) and OCR analysis technologies, integrates the extracted drawing-based information with the building use, gross floor area, structure, and site area stored in the building administration system (DB) into the input prompt of the LLM-based judgment module, performs drawing-based area calculation, determines whether routes are separated, checks for obstructions in emergency passages, determines whether restricted areas under other laws regarding public open spaces, light areas, evacuation passages, road occupancy areas, and mandatory parking areas are included, and identifies conditions requiring confirmation by departmental personnel, and controls the generation and transmission to the user terminal of warning data based on the results identified by the judgment module, which displays the possible outdoor business area and scope on the drawing when permitted, and includes the name of the violating statute, article number, grounds, and contact information of the relevant department when restricted; the control unit also controls the direct accessibility between the entrance and the outdoor business area on the drawing, the possibility of movement between floors, and the possibility of contiguous connections within the site, An AI-based automatic determination of outdoor business eligibility targets and licensing verification system characterized by determining whether other use spaces intervene in common spaces, corridors, and parking lots according to pre-established objective criteria, quantifying the possibility of inclusion in restricted areas by calculating the overlap ratio and center coordinate error range with restricted areas under other laws, and generating a structured restriction condition dataset that classifies items requiring departmental review corresponding to each restriction condition and can be transmitted together with the judgment results. Claim 4 In paragraph 3, the control unit identifies whether the location of the building or outdoor business premises falls under any one of the following: public open space, open space within the site, evacuation passage, road occupancy area, or mandatory parking area by matching it with a previously stored database of restriction conditions under other laws based on location information, floor plan, and layout data of the building; if it falls under the restriction area, it generates warning data including the name of the relevant law, article number, name of the responsible department, and contact information for each item where there is a risk of violation; transmits the warning data to the electronic signature module of each relevant department to collect signatures; and controls the issuance of a permit verification result report to the user terminal once signatures from all relevant departments are completed. This characterizes an AI-based automatic determination of eligible outdoor business premises and permit verification system. Claim 5 In claim 4, the control unit calculates statistical similarity and pattern matching rates for each object by comparing the location coordinates and attribute data of an object extracted from the drawing file with first historical data corresponding to the same industry, similar buildings, and similar districts among the past permit verification result log data accumulated in the storage unit, and classifies the building as a subject for in-depth review when the similarity and pattern matching rates are each below a preset threshold.

Images