Dialogue intelligent agent interaction method and system for LED display screen industry

Abstract

The application relates to the field of computer software and artificial intelligence technology, and provides a dialogue intelligent agent interaction method and system for the LED display screen industry.The method receives text, voice, pictures, file attachments and authorized position input through a graphic client, and normalizes the input into an item parameter structure with field state and geographical position context, executes follow-up questions and state access control according to the required field set of intention and target business capability, calls a bill of materials, a quotation, a drawing, a scheme or a regional service routing engine, searches a module, a light emitting device, a chip, a printed circuit board, a power supply, a box, a control system and industry technical evidence, jointly determines a business engine answer or matching data, isolates data access according to an authenticated tenant context, does not output a completion prompt when the business engine is not successful, and delivers a file only after the source, adaptive parameters and permission verification are passed and a unique target file is determined.

Description

Technical Field

[0001] This application relates to the fields of computer software and artificial intelligence technology, and in particular to a dialogue agent interaction method and system for the LED display industry. Background Technology

[0002] Currently, existing solutions in the field of AI-assisted design in the LED display industry mainly fall into three categories: general-purpose AI dialogue tools, traditional Software as a Service (SaaS) platforms, and general-purpose intelligent agent frameworks. General-purpose AI dialogue tools lack LED industry expertise and cannot access business systems such as Bill of Materials (BOM) engines, Computer-Aided Design (CAD) parametric generation, and price databases. Traditional SaaS platforms are organized independently into modules such as display simulation, BOM, CAD, quotation, and proposal, requiring users to switch between multiple pages. General-purpose intelligent agent frameworks are resource-intensive and unsuitable for lightweight deployment as mini-programs.

[0003] Because tasks such as screen calculation, bill of materials, pricing, computer-aided design, proposals, and end-number matching are scattered across different modules, users need to repeatedly switch pages and enter information repeatedly. The four input methods—text, voice, images, and files—cannot be integrated into a unified project parameter structure. Due to the time limit of a single request in the mini-program, long-running tasks lack progress feedback. After the bill of materials is generated, it cannot automatically match quotations, and artificial intelligence may fabricate prices and inventory. In addition, the existing system cannot jointly calculate the project address or authorized location with the service provider's coverage area, service radius, capabilities, and qualifications. After the quotation or drawings are completed, manual searching for installation and after-sales resources is still required.

[0004] Existing Q&A products in the display industry often separate document retrieval from engineering calculations: when generating answers based solely on similar text, it's impossible to determine the receiving card's load capacity, controller port, module scanning method, and configuration file compatibility; when sending materials solely by product name, it may send incorrect regions, languages, versions, or incompatible configuration files. Existing solutions typically lack a mechanism to bind document source, version, page number, or content location to engineering rule calculation results, and also lack status gates to block transmission when the target file cannot be uniquely identified according to preset rules, the source is untraceable, or there is no delivery authorization. Summary of the Invention

[0005] Therefore, it is necessary to provide a dialogue intelligence interaction method and system for the LED display industry to solve the problems in the existing technology, such as the lack of a unified dialogue entry point for multiple business systems, fragmented multimodal input methods, lack of progress feedback for long-time business, the inability to automatically link the bill of materials and price database, the inability to automatically route regional installation and after-sales resources according to project location, the possibility that tenant identity may be forged by the client, and the possibility that artificial intelligence may fabricate price inventory or service provider data.

[0006] Firstly, this method provides a dialogue agent interaction method for the LED display industry. The method receives multi-source information input by the user within a dialogue window of a graphical client. This multi-source information includes one or more of text, voice, images, file attachments, and user-authorized location information. It verifies the login credentials, session token, or platform identity carried by the client, determines the tenant identifier on the server side, and forms an authenticated tenant context. Normalization processing is performed on the multi-source information, uniformly extracting different types of multi-source information into a project parameter structure, and converting the location information into a geographic location context within the project parameter structure. Intent recognition is performed on the project parameter structure using a semantic model, extracting project parameter fields from the multi-source information, maintaining a state machine for the project parameter structure including field source, confirmation status, and version, and generating structured follow-up cards within the dialogue window to receive supplementary information from the user when the project parameter structure lacks key fields. Based on the intent recognition results, a target business capability is determined from multiple LED display industry-specific business capabilities and applied to the project parameter structure. When the set of required fields corresponding to the target business capability in the data structure meets the preset state access control, the corresponding business engine is invoked to perform business calculations, and the progress and results of the business calculations are fed back through status cards in the dialog window; when the business calculation results include a bill of materials, the price library engine is invoked to match the quotations of each material, the quotations of the materials that match the quotations are filled in, and the materials that do not match the quotations are marked as pending quotations. All data access during the business calculation process carries the authenticated tenant context; when the user intent involves installation, after-sales service, maintenance, inspection, spare parts delivery or regional service, the regional service routing engine is invoked to filter and sort candidate service resources according to the geographical location context; the storage space of business data, files and audit records is isolated according to the authenticated tenant context, and the business calculation results and service matching results are archived to the isolated storage space of the current tenant; when the successful execution result of the business engine is not received, the semantic model does not output a prompt message that the business has been completed.

[0007] In implementations of Q&A, product selection consultation, fault diagnosis, or data acquisition in the display industry, the system calls the industry knowledge retrieval engine to form an evidence set with source information based on the project parameter structure; for issues involving load capacity, compatibility, connection topology, or configuration adaptation, the system orchestrates the participation of the deterministic business engine in the calculation; for data acquisition requests, the system calls the data matching engine to perform model, region, language, version, adaptation parameters, and access permission verification, and generates a file delivery card after determining a unique target file from multiple candidate files, either when there is only one candidate file or based on preset version priority, validity, and adaptation rules.

[0008] Secondly, a dialogue agent interaction system for the LED display industry is provided. The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the dialogue agent interaction method for the LED display industry described above.

[0009] Thirdly, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the aforementioned dialogue agent interaction method for the LED display industry.

[0010] Fourthly, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described dialogue agent interaction method for the LED display industry.

[0011] This application unifies multiple independent business systems within a single dialog window, enabling users to complete end-to-end design, quotation, and procurement assistance for display projects using natural language, without switching between different pages and modules. By normalizing and extracting text, voice, images, and file attachments into a unified project parameter structure, the application eliminates the problem of fragmented input methods. Intent recognition and structured follow-up cards complete key fields when parameters are missing, preventing AI from directly fabricating missing information. Automatic linkage between the bill of materials (BOM) and price database automatically matches quotations for each material after the BOM is generated and marks unmatched items as pending quotations, preventing AI from fabricating price inventory. Data access is constrained by authenticated tenant context, and installation or after-sales service resources are jointly screened based on geographic location context, service radius, service capabilities, and qualifications; if no match is found, a "no service available" status is returned. This application simultaneously solves the problems of scattered multiple business systems, fragmented input methods, lack of progress feedback for long tasks, and inability to automatically match prices, achieving integrated delivery of end-to-end design, quotation, and procurement assistance.

[0012] By incorporating the industry knowledge retrieval engine, deterministic business engine, and data matching engine into the same dialogue orchestration chain, knowledge responses no longer rely solely on semantic model generation but are bound to document sources and engineering calculation results. By performing model, region, language, version, adaptation parameters, and permission checks on configuration files and product data, it is possible to prevent incorrect versions, incorrect configurations, or unauthorized files from being sent, and to make each response and document delivery traceable based on audit records. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] in: Figure 1 This is a flowchart of a dialogue agent interaction method for the LED display industry, as described in this application. Figure 2 This is a schematic diagram of the module of the dialogue intelligent agent interaction system for the LED display industry in the embodiments of this application; Figure 3 This is a structural block diagram of the server computer device in the embodiments of this application; Figure 4 This is a structural block diagram of the client computer device in an embodiment of this application. Detailed Implementation

[0015] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0016] Please see Figure 1 As shown, Figure 1 A flowchart illustrating the dialogue agent interaction method for the LED display industry provided in this application embodiment includes the following steps: Step 01: Receive multi-source information input by the user in the dialog window of the graphical client. The multi-source information includes one or more of the following: text, voice, images, file attachments and location information authorized by the user. Verify login credentials, session tokens or platform identity identifiers. Determine the tenant identifier and the authenticated tenant context on the server side. Step 02: Perform normalization processing on multi-source information, and extract different types of multi-source information into a project parameter structure with field source, field status, version and geographic location context; Step 03: Perform intent recognition on the project parameter structure through semantic model, extract project parameter fields from multi-source information, maintain the project parameter structure state machine including field source, confirmation status and version, and generate structured follow-up question cards in the dialog window to receive supplementary information from the user when the project parameter structure is missing key fields. Step 04: Based on the intent recognition results, determine the target business capability from multiple LED display industry-specific business capabilities. When the set of required fields corresponding to the target business capability in the project parameter structure meets the status access control, call the corresponding business engine to perform business calculations and provide feedback on the progress and results of the business calculations through status cards. Step 04 above may include: Step 04A: When the target business capability is industry knowledge Q&A, product selection consultation or fault diagnosis, the industry knowledge retrieval engine is invoked to form an evidence set; when the problem involves engineering calculation or compatibility judgment, the deterministic business engine is invoked, and the retrieval evidence and calculation results are jointly packaged into a knowledge answer card. Step 04B: When the target business capability is data acquisition, call the data matching engine to filter candidate files, and perform model, region, language, version, adaptation parameters and permission verification. If there is only one candidate file, or if multiple candidate files determine a unique target file according to preset rules, a file delivery card is generated; if there are multiple candidates and a unique target file cannot be determined according to the rules, a selection card is generated, and if the verification fails, the corresponding incomplete status is returned. Step 05: When the business calculation result includes a bill of materials, call the price library engine to match the price of each material, fill the price information of the materials with matching prices, and mark the materials without matching prices as pending inquiry status. All data access during the business calculation process carries the authenticated tenant context. Step 06: Under the service intent, invoke the regional service routing engine based on the geographic location context; isolate the storage space of business data, files and audit records according to the authenticated tenant context, and archive the business calculation results and service matching results; Step 07: If the semantic model does not receive a successful execution result from the business engine, it will not output a message indicating that the business has been completed.

[0017] Specifically, in step 01, voice input is recorded via the WeChat Mini Program recording interface (wx.startRecord) and then transcribed into text using the Automatic Speech Recognition (ASR) service from iFlytek or Tencent Cloud. Image input extracts text information through the Optical Character Recognition (OCR) module and identifies scene visual features such as the installation environment, the area to be installed, wall material, and brightness conditions through a visual encoder or multimodal visual model. In one implementation, after the user takes a photo of the scene environment, the OCR extracts text information such as "indoor, shopping mall, 4.5-meter ceiling height, hoisting," where "4.5-meter ceiling height" is written into the "installation environment" field of the project parameter structure, and "hoisting" is written into the "installation method" field. Location information is obtained from user-authorized platform location interfaces, manually entered project addresses, or addresses in project files, and is standardized as latitude and longitude, administrative region, location accuracy, address source, and authorization status. File attachments support comma-separated value format, spreadsheet format, and spreadsheet format with macros enabled. After parsing the table content using the price database importer, the file enters the draft review process. Tenant identifier format validation uses regular expressions, consisting of uppercase and lowercase letters, numbers, underscores, and hyphens, with a length of 3 to 64 characters. Service is rejected if validation fails.

[0018] In step 02, the project parameter structure includes fields such as project name, customer, installation location, screen width and height, usage environment, pixel pitch, module specifications, cabinet specifications, installation method, edging, back frame, steel material, structure type, profit margin, field source, field status, version, and geographic location context. In one implementation, the user inputs "make a 4×3 meter P2.5 indoor screen" via voice. The system extracts the screen width as 4000 mm, the screen height as 3000 mm, the pixel pitch as 2.5 mm, and the usage environment as indoor. The user then takes a photo and uploads a picture of the on-site installation environment. OCR recognizes "shopping mall, 4.5 meters high" from the photo. The system adds "4.5 meters high" to the "installation environment" field of the project parameter structure and "shopping mall" to the "application scenario" field. The size information provided by voice and the environmental information provided by the image are superimposed on the same project parameter structure. Each field is organized according to a preset data pattern, allowing the downstream semantic model and business engine to read parameters from the same set of fields.

[0019] In step 03, intent recognition uses the Bill of Materials module, Control System Selection module, Screen Pricing module, Drawing Generation module, Proposal Generation module, Inventory Matching module, Regional Service module, Industry Knowledge Q&A module, and Controlled Data Delivery module to route user messages to the corresponding business modules. The state machine of the project parameter structure tracks the source, version, and confirmed, pending, default, or deduced status of each field, and executes state access control according to the set of required fields for the target business capability. The structured follow-up question card contains one or more sets of fields to be confirmed, each set containing field labels, a candidate list, and a manual input entry. In one implementation, the user inputs "Make me a P2.5 screen," the system recognizes that the screen width and screen height fields are missing, generates a decision card, and the card contains the "Screen Size" field and candidate options "4×3 meters, 6×4 meters, 8×5 meters, manual input." After the user clicks "4×3 meters," the system writes the value into the confirmation record of the project parameter structure.

[0020] In step 04, several LED display industry-specific business capabilities include bill of materials calculation capability, control system link planning and equipment selection capability, screen quotation capability, drawing generation capability, solution generation capability, inventory matching capability, regional installation, after-sales or service routing capability, industry knowledge Q&A capability, and controlled data delivery capability. The bill of materials engine reads parameters from the project parameter structure: width 4000 mm, height 3000 mm, pixel pitch 2.5 mm, usage environment indoors, and installation method wall-mounted. It calculates the number of module columns as 4000 divided by 320 equals 12.5, rounded up to 13 columns, and the number of module rows as 3000 divided by 160 equals 18.75, rounded up to 19 rows. The total number of modules is 13 multiplied by 19 equals 247, corresponding to an actual screen width of 4160 mm and an actual screen height of 3040 mm. When the actual screen size is inconsistent with the target size input by the user, the actual screen size is marked as pending confirmation, and the bill of materials, quotation, drawing, or proposal generation engine is invoked only after user confirmation. During business computation, short tasks return results synchronously and directly, medium-sized tasks return intermediate results in chunks via streaming responses, and long tasks initiate background tasks and return task identifiers. Progress information is fed back to the dialog window through at least one event channel, including server-sent events, WebSocket, chunked responses, polling, or message queue callbacks. Results are returned upon completion of the computation. WebSocket is a network protocol for full-duplex (bidirectional) communication over a single TCP connection.

[0021] When a user requests a control system solution, the control system selection engine reads the confirmed module row and column layout, overall screen resolution, screen dimensions, input signals, transmission distance, and redundancy requirements. It then generates candidate load sub-areas and directions for receiver cards, determines the number of receiver cards, transmitting ports or optical ports, and cables based on audited specifications, and generates candidate links between the fixed video processing equipment and the modular splicing equipment chassis and input / output board combinations. After each candidate link is verified for load, port, slot, interface, and primary / backup relationship, a control system solution confirmation card is returned. Unconfirmed inferred parameters are not directly included in the bill of materials and quotation.

[0022] In one implementation, the user says in the dialog "Generate project delivery documents according to the current project". The system detects that the estimated time is about 55 seconds, returns a task identifier and returns within 0.5 seconds. Then, at the 2nd second, it pushes "Progress 10%, Loading project parameters", at the 10th second, it pushes "Progress 40%, Business processing in progress", at the 30th second, it pushes "Progress 70%, Rendering output", at the 50th second, it pushes "Progress 95%, Generate preview image", and at the 55th second, it pushes "Progress 100% complete and file download address".

[0023] In step 05, the bill of materials (BOM) output by the BOM engine includes material category, specifications, quantity, and supplier name. The price database engine queries by category. For materials with an active status, the supplier name, price, inventory, and discount are read from the price database and populated into the quotation field of the BOM. Materials without a match are marked as "Pending Quotation" in the dialog window. In one implementation, the BOM contains 247 P2.5 indoor modules (320×160 mm). After the price database engine matches this specification, it populates a quotation of 37 yuan per module, totaling 9139 yuan, and marks it as "Matched". Another material model does not exist in the price database, is marked as "Pending Quotation", and the "Pending Quotation" flag is displayed in the dialog window. All data access carries the tenant identifier, and only data of the current tenant is returned.

[0024] In step 06, the tenant identifier is determined by the server based on login credentials, session tokens, or platform identity identifiers; business data, files, and audit logs can be logically or physically isolated through independent directories, object storage namespaces, database schemas, table partitions, or row-level tenant keys with mandatory access control, and cross-tenant access is denied without exposing other tenant data.

[0025] In step 07, if the bill of materials engine does not return the calculation result, the semantic model must not output the "generated" prompt. Instead, it should only output the prompt that the service is temporarily unavailable and retain the complete requirements currently entered by the user. The user can continue to execute the process after the business engine is restored.

[0026] Through the above steps, multi-source inputs converge into a unified project parameter structure. End-to-end delivery is achieved by scheduling business systems such as bill of materials (BOM) accounting, control system link planning and equipment selection, screen quotation, drawing generation, solution generation, inventory matching, regional service routing, industry knowledge Q&A, and controlled data delivery through a unified dialogue entry point. Real-time linkage between the BOM and price database automatically matches quotations after BOM generation and marks unmatched items as pending quotations, preventing the semantic model from fabricating price data. Real-time progress feedback is provided within the time limit of a single request in the mini-program through long-task streaming progress pushes. The regional service routing engine is driven by geographic location context to match installation, after-sales, maintenance, inspection, or spare parts delivery resources, and returns a "no service available" status when no matching candidates are found. Real-state access control constraints prevent the semantic model from outputting a completion status prompt when the business engine fails to execute.

[0027] like Figure 2 As shown, this application also provides a dialogue-based intelligent agent interaction system for the LED display industry, with a five-layer system architecture: The first layer (L1) is a lightweight graphical client layer, typically implemented using the WeChat Mini Program platform. It includes the dialogue homepage (ai-home.js), the price library page (price-library), and the AI ​​design page (ai-design). The dialogue homepage contains a dialogue message flow component and a message input area component. The dialogue message flow component renders various message cards, including plain text message cards (text bubbles), structured follow-up question cards (decision-card), intelligent computing progress cards (quote-card), CAD generation progress cards (drawing-card), proposal generation cards (proposal-card), tail inventory matching cards (tail-card), price import result cards (price-result-card), CAD task progress cards (cad-status-card), and delivery document cards (artifact-result). The message input area component includes text input boxes, a voice recording button, a file attachment upload button, and a send button.

[0028] The lightweight graphical client can also be used for Alipay mini-programs, Douyin mini-programs, H5 web applications, progressive web applications, hybrid applications, React Native applications, Flutter applications, desktop clients, or self-built applications, third-party applications, workbench web applications, customer contact tools, or intelligent robots in enterprise collaboration platforms; different clients all present message flows and structured status cards through graphical dialogue windows, application messages, interactive cards, or robot messages.

[0029] In the implementation of the enterprise collaboration platform, the channel adapter receives messages or events pushed by the platform, performs signature verification and encrypted message decryption on callback requests, and performs deduplication based on at least one of event identifier, message identifier and idempotent key; the channel adapter maps enterprise identifier, application identifier, member identifier or external contact identifier to authenticated tenant context, and associates platform sessions with project parameter structures, engineering projects or business tasks.

[0030] For example, in the WeChat Work implementation, internal members or external contacts can submit requests for selection, quotation, drawings, solutions, installation, or after-sales service through self-built applications, third-party applications, workbench web pages, customer contact tools, or intelligent robots. After the system completes the parameter status access control, it calls the corresponding LED display industry business engine and returns the task progress and results through application messages, interactive cards, robot replies, or file links. When the request involves installation, after-sales service, or maintenance, the project address, authorized location, or customer address is written into the geolocation context, and the regional service routing engine matches service resources and generates executable appointments, dispatch orders, or transfers to manual operations.

[0031] The second layer (L2) is the intelligent routing layer, which runs on cloud functions (cloudfunctions / ai-router). It includes a unified entry function index.js and multiple capability adaptation routes. These capability adaptation routes correspond to one or more of the following business capabilities: bill of materials accounting, control system link planning and equipment selection, computing screen quotation, drawing generation, solution generation, inventory matching, regional services, industry knowledge Q&A, and controlled data delivery.

[0032] The third layer (L3) is the business engine layer, which includes the BOM engine, control system selection engine, CAD engine, price database engine, quotation engine, proposal engine, and regional service routing engine. The regional service routing engine filters and sorts candidate service providers based on geographical context, service radius, LED display type capabilities, qualifications, response time, ratings, and tenant preferences. The BOM engine calculates the number of modules, enclosures, control devices, and cables after receiving project structural parameters. The CAD engine parametrically generates DXF format computer-aided design files after receiving steel structure parameters. The price database engine queries supplier, price, inventory, and discount information for each material after receiving the bill of materials. The quotation engine calculates costs and profit margins after receiving the bill of materials and price database query results. The proposal engine populates templates and merges customer information after receiving project parameters and calculation results.

[0033] The control system selection engine is used to jointly generate candidate control links consisting of receiving cards, transmitting devices or video processing devices, modular splicing devices, input / output boards, fiber optic transceivers and cables based on the screen's geometric layout and control requirements, and outputs port allocation, connection topology, redundancy relationships and verification status.

[0034] The fourth layer (L4) is the industry knowledge layer, which stores a knowledge base specifically for the display industry, including more than 35 types of modules, more than 37 types of cabinets, more than 67 types of control devices, more than 702 compatibility relationships, 10 types of ontology classifications, and 4 engineering rules. It adopts a two-layer architecture of industry general layer and enterprise brand layer, with data written in isolation in the enterprise brand layer.

[0035] Industry knowledge is not limited to control system knowledge. Its objects include LED display modules, LED chips or other light-emitting devices, driver chips, printed circuit boards, power supplies, enclosures, control system equipment, steel structures and installation materials, software, and configuration files. Knowledge content includes product naming and specifications, material composition, interfaces and compatibility, selection and load-bearing rules, manufacturing processes, installation and maintenance, failure cases, safety regulations, industry standards, product lifecycles, version changes, and industry technology trends. When industry technology trends are included as knowledge objects, they are linked to the publication time, scope of application, and source level; unverified marketing information should not be substituted for technical evidence.

[0036] Each evidence unit in the industry knowledge layer is associated with at least one or more metadata elements, including the content text, document identifier, document version, source location, applicable product, publication or update time, validity status, source level, and tenant access permissions. During retrieval, keyword retrieval, vector semantic retrieval, product model field retrieval, and knowledge graph relationship retrieval can be combined, and candidate evidence can be rearranged according to source level, version validity, product applicability, and field consistency.

[0037] In one implementation, when a user asks "How to configure a 6500mm x 4500mm, P2.5, 320mm x 160mm module using a certain brand of control system", the system first retrieves valid specification evidence of the receiver card, controller, and interface from the product specification library. Then, it calls the control system selection engine to calculate the full screen resolution, receiver card carrying sub-area, number of ports, and connection topology. Finally, the system marks the document facts, calculation results, and parameters to be confirmed in the response, and attaches the document name, version, and page number or content location.

[0038] In another implementation, when a user inquires about the appropriate LED chips, driver chips, PCB layer count, and power supply for a given module specification, the system retrieves module specification evidence, device electrical parameters, driving methods, current and power consumption rules, PCB constraints, and power supply load conditions. When the question involves power consumption, current margin, or the number of power supplies, the corresponding calculation processor is invoked to present product evidence, rule calculation results, and uncertainties. For issues related to enclosure heat dissipation, installation materials, construction and maintenance, troubleshooting, and changes in industry standards or product versions, the same source binding and real-state access control are used.

[0039] In another implementation, a user requests "send the configuration file corresponding to the P2.5, 320mm x 160mm module and the specified receiver card". The data matching engine reads the module pixels, scanning method, driving conditions, interface type, hub board, and receiver card adaptation metadata of the candidate files from the file library. If the above parameters are verified and the current tenant has delivery permissions, a file delivery card is generated when there is only one candidate file, or when multiple candidate files determine a unique target file based on preset version priority, validity, and compatibility rules. If only P2.5 is considered and a unique target configuration file cannot be determined according to the rules, the system generates a selection card requiring supplementary scanning method, driver chip, or interface information, and sends the file ambiguously without adhering to the pixel pitch.

[0040] For product specifications, user manuals, or software installation packages, the data matching engine further filters candidate files based on product series, domestic or overseas region, Chinese or foreign language, hardware version, software version, and expiration date. For data sources that can only read directories or indexes but lack original file delivery permissions, they are marked as pending source leads, preventing the engine from bypassing permissions to obtain or send the original files. Each response and delivery record includes session identifier, tenant identifier, file identifier, file version, source, matching fields, validation results, and sending time.

[0041] The fifth layer (L5) is the data storage layer, which is logically or physically isolated according to the authenticated tenant context. In one directory isolation implementation, the directory structure is data / tenants / {tenant_id} / , which includes the price library directory (supplier-library-v2 / libraries / ), the project data directory (project / ), the service provider library directory (service-providers / ), the file directory (files / ), and the audit log directory (logs / ).

[0042] The following describes the dialogue agent interaction method for the LED display industry provided in this application with reference to specific embodiments: In some implementations, multiple display industry-specific business capabilities include bill of materials accounting capabilities, control system link planning and equipment selection capabilities, display quotation capabilities, drawing generation capabilities, solution generation capabilities, inventory matching capabilities, regional installation, after-sales or service routing capabilities, industry knowledge Q&A capabilities, and controlled data delivery capabilities. There is a real-time linkage between the bill of materials engine corresponding to the bill of materials accounting capability and the price database engine corresponding to the bill of materials pricing capability.

[0043] Specifically, the bill of materials (BOM) accounting capability is executed by the BOM engine, which calculates the quantity of modules, enclosures, control devices, and cables after receiving the project's structural parameters. Regional installation, after-sales, or service routing capability is executed by the regional service routing engine, which, after receiving the geographic location context and service type, filters and sorts candidate service resources from the service provider database currently available to the tenant. The screen pricing capability is executed by the pricing engine, which calculates costs and profit margins after receiving the BOM and price database query results. The drawing generation capability is executed by the Computer-Aided Design (CAD) engine, which parametrically generates CAD design documents after receiving steel structure parameters. The proposal generation capability is executed by the proposal engine, which populates templates and merges customer information after receiving project parameters and calculation results. The inventory matching capability is executed by the tail engine, which matches certified supplier inventory, prices, and discount information after receiving the BOM.

[0044] The real-time linkage between the Bill of Materials (BOM) engine and the Price Database engine is as follows: After the BOM engine outputs the BOM, the system automatically calls the Price Database engine, which queries according to material category. In one implementation, the BOM engine outputs 247 sheets of P2.5 indoor modules (320×160 mm). The system automatically calls the Price Database engine to query this type of material. Once the Price Database engine finds an active material, it fills in a price of 37 yuan per sheet, totaling 9139 yuan, and marks it as "Matched". Another material model does not exist in the price database, so the Price Database engine does not find it. This material is marked as "Pending Inquiry," and the "Pending Inquiry" flag is displayed in the dialog window, and an inquiry task is automatically generated. The material categories queried by the Price Database engine include seven categories: modules, power supplies, control equipment, enclosures, cables, steel structures, and accessories. Each category matches a corresponding supplier database and price item. Only materials in an active state are filled with a price; materials in draft state are not included in the pricing process.

[0045] In some implementations, receiving multi-source information input by the user within a dialog window includes: It receives voice input and transcribes it into text using an automatic speech recognition service; Receive image input and extract text and visual features from the image input using optical character recognition service; Receive file attachment input; file attachments support various table formats.

[0046] Specifically, voice input is recorded via the WeChat Mini Program recording interface (wx.startRecord) and then transcribed into text using iFlytek or Tencent Cloud's Automatic Speech Recognition (ASR) service. Image input extracts text information through an optical character recognition module and identifies scene visual features such as brightness, floor height, wall material, and the area to be installed through a visual encoder or multimodal visual model.

[0047] In one implementation, after the user takes photos of the site environment, OCR recognizes textual and visual features such as "indoor, shopping mall, 4.5-meter ceiling height, hoisting." "4.5-meter ceiling height" is written into the installation environment field of the project parameter structure, and "hoisting" is written into the installation method field. Document attachments are parsed using the price library importer (SupplierLibraryService.create_import) to parse comma-separated value formats, spreadsheet formats, and spreadsheet formats with macros enabled. The parsed material data enters the draft review process and does not directly enter the official price library.

[0048] In some implementations, the project parameter structure includes project name, customer, installation location, screen width and height, usage environment, pixel pitch, module specifications, cabinet specifications, installation method, edging, back frame, steel material, structure type, profit margin, field source, field status, and version fields, and includes a geographic location context consisting of one or more of the following: project address, latitude and longitude, administrative region, positioning accuracy, address source, and user authorization status.

[0049] Specifically, the project parameter structure contains fields corresponding to all the core parameters required for an LED display project from requirements gathering to quotation delivery. Project name, customer, and installation location identify the project's ownership and implementation location. Screen width and height are used to calculate the number of module columns, module rows, and the total number of modules. The number of module columns is obtained by dividing the screen width by the module width and rounding up; the number of module rows is obtained by dividing the screen height by the module height and rounding up. When the actual screen size calculated based on the rounded-up result differs from the target size input by the user, the actual screen size, size deviation, and its source are written into the project parameter structure and marked as pending confirmation. After user confirmation, the corresponding field is allowed to pass through the status gate and the subsequent business engine is invoked.

[0050] The usage environment distinguishes between indoor and outdoor scenarios, influencing module model selection (high-brightness models are required for outdoor use). Pixel pitch determines display resolution, module specifications determine the physical size of a single module, and cabinet specifications determine the size of the mounting carrier. The installation method determines the steel structure parameter template used by the computer-aided design engine; the edging and back frame determine the structural boundary design in the computer-aided design drawings; the steel material determines the material selection in steel structure calculations; and the structure type determines the mechanical calculation model in the steel structure design. Profit margin is used by the pricing engine to calculate the final price.

[0051] In some implementations, the corresponding business engine is invoked to perform business calculations, including: When the estimated time for business calculation is less than a preset first threshold, the result is returned synchronously; When the estimated time for business calculation is greater than or equal to a preset first threshold and less than a preset second threshold, intermediate results are returned in blocks using streaming response. When the estimated time for business calculation is greater than or equal to the preset second threshold, a background task is started and a task identifier is returned. Progress information is pushed to the dialog window through the server's event channel, and the result is called back after the calculation is completed.

[0052] Specifically, the first threshold corresponds to 5 seconds, and the second threshold corresponds to 30 seconds. When the estimated computation time is less than 5 seconds, the result is returned synchronously directly. When the estimated time is between 5 and 30 seconds, an intermediate result is returned via at least one event channel, including server-sent events, WebSocket, chunked responses, polling, or message queue callbacks. When the estimated time exceeds 30 seconds, a background task is started and a task identifier is returned immediately. The user interface receives the task identifier within 0.5 seconds, and then progress events are fed back via event channels. A structured event is pushed every 2 to 5 seconds, containing percentage progress, current step name, and status message. Upon completion of the calculation, a completion event is pushed, containing a status of "completed," the file download address, and the preview address.

[0053] In one implementation, a user requests the generation of project delivery documents, which is expected to take approximately 55 seconds. The system returns a task identifier at 0.5 seconds, pushes a progress bar of 10% and a status indicating project parameters are loading at 2 seconds, pushes a progress bar of 40% and a status indicating business processing is in progress at 10 seconds, pushes a progress bar of 70% and a status indicating rendering output at 30 seconds, pushes a progress bar of 95% and a status indicating preview image generation at 50 seconds, and pushes a progress bar of 100% and a completion event including the file download address at 55 seconds. The user can click the "Stop Generation" button at any time during task execution, and the system retains the confirmed project parameters and deducts the used points. This long task management mechanism solves the problem of long-running business processes not returning results in a timely manner under the single request time limit of lightweight clients or gateways.

[0054] In some implementations, the status card types include plain text message cards, structured follow-up question cards, screen rendering progress cards, drawing generation progress cards, proposal generation cards, inventory matching cards, price import result cards, and regional service matching cards. Structured follow-up question cards contain one or more sets of fields to be confirmed. Screen rendering progress cards contain parameter completeness and screen preview information. Drawing generation progress cards contain rendering progress and layer management information. Proposal generation cards contain chapter options and customer information. Inventory matching cards contain supplier, inventory quantity, price, and discount information. Price import result cards contain pass / fail statistics, failure statistics, and error details. Regional service matching cards contain service provider name, service type, distance, service radius, response time, qualification level, rating, availability status, and executable operations.

[0055] Specifically, plain text message cards are used for plain text conversations. Structured follow-up question cards are used for structured completion when parameters are missing. They contain one or more sets of fields to be confirmed, each set including field labels, a list of candidate options, and a manual input field. Users can complete the field completion by clicking on the candidate options or manually entering the information.

[0056] The calculation progress card is used for progress feedback during the bill of materials calculation and calculation quotation process. It includes parameter completeness (showing the percentage of confirmed fields out of all required fields) and screen preview information (showing a preview of the module layout).

[0057] Drawing generation progress cards are used to provide progress feedback during the computer-aided design (CAD) file generation process, including rendering progress (current generation progress percentage) and layer management information (showing the number of layers that have been completed).

[0058] The proposal generation card is used for progress feedback during the proposal generation process. It includes chapter options (users can select chapters such as project overview, design scheme, material details, construction scheme, etc.) and customer information (displaying confirmed fields such as customer name, project name, and installation location).

[0059] The inventory matching card is used to display the results of the end-of-line inventory matching, including supplier name, inventory quantity, price and discount information.

[0060] The Price Import Results Card provides feedback on the price import results for document attachments. It includes pass / fail statistics, failure statistics, and error details. The error details list the reasons for failure line by line (such as missing cost price, missing supplier name, invalid multiplier, etc.).

[0061] In some implementations, the price import result card is generated through the following steps: Parse the file attachment; Verify the cost price, specifications, and supplier name of each material line by line; Place the validated materials into draft mode; The materials that failed the verification are summarized into error details; Upon receiving the user's activation request, the draft material will be switched to the activated state.

[0062] Specifically, when parsing file attachments, the price library importer (SupplierLibraryService.create_import) is called to read each line of data from a comma-separated value format or spreadsheet format file. Each line is validated to ensure the three required fields for each material—cost_price, specification (spec), and supplier name (supplier_name)—are complete, and the quote multiplier is also verified to be a valid value. Material records that pass validation are marked as drafts and added to the tenant's price library directory but are not used for quote queries. Materials that fail validation are summarized into an error details array, with each record's line number and the reason for failure (e.g., missing cost price, missing supplier name, invalid quote multiplier, etc.).

[0063] After the user clicks the activation button on the front end, the system calls the activation interface (activate_import) to switch all draft materials to active status. At this point, the materials officially enter the price database and participate in subsequent quotation matching queries. Draft materials are not entered into the quotation process before activation to ensure that unverified supplier prices are not mistakenly used in project quotations.

[0064] In some implementations, the method further includes a real-state access control step: When the regional service routing engine does not match a candidate service provider that meets the service radius, capability, qualification or tenant authorization conditions, it outputs a state of no available service and does not fabricate service provider, distance, qualification, price or reservation status. When no quote is matched in the semantic model, output the status of pending quote without fabricating supplier, price, inventory, discount, or delivery information; When the business engine is unavailable, output a message indicating that the service is temporarily unavailable while retaining user requests.

[0065] Specifically, when the semantic model does not match a quote, the price database engine returns an empty result, and the semantic model outputs a pending quote status and displays a pending quote indicator in the dialog window, without fabricating supplier names, price values, inventory quantities, discount percentages, or delivery date information. In one implementation, if a certain model of power supply in the bill of materials does not exist in the price database, the semantic model outputs "Pending quote, no matching supplier available," instead of fabricating false information such as "a supplier quoted XX yuan."

[0066] When the business engine is unavailable (e.g., the bill of materials engine service is interrupted, the computer-aided design engine times out, or the price database engine connection fails), the semantic model outputs a message indicating that the service is temporarily unavailable. Simultaneously, it saves the user's complete current requirements (including confirmed project parameter fields, the user's original input text, and the current execution status) in the isolated task state storage space of the current tenant. After the business engine recovers, the user can continue executing unfinished tasks from the saved state without having to re-enter the provided information. In one implementation, when a user requests the generation of computer-aided design drawings, the computer-aided design engine service is interrupted. The semantic model outputs "CAD generation service is temporarily unavailable, please try again later," and saves the user's current screen width (4000 mm), height (3000 mm), pixel pitch (2.5 mm), and installation method (wall-mounted) parameters in the isolated task state storage space. When caching on the client side, the cached data is encrypted and an expiration date is set. After recovery, the user can continue by clicking "retry" without having to re-enter the aforementioned parameters.

[0067] In some implementations, the tenant identifier is determined by the server based on login credentials, session tokens, or platform identity identifiers, and format and authorization relationship verification is performed. Business data, files, and audit records are logically or physically isolated based on tenant identification using at least one of the following methods: independent directory, object storage namespace, database schema, table partitioning, or row-level tenant key with mandatory access control. Bind the price database, product database, service provider database, file database, and audit logs to tenant identifiers, tenant storage namespaces, and access control policies; Cross-tenant data access returns an explicit error and does not expose other tenants' data.

[0068] Specifically, the tenant identifier is determined by the server based on login credentials, session tokens, or platform identity identifiers, and the server verifies the authorization relationship between the identity and the tenant. The format of the tenant identifier is validated using the regular expression [A-Za-z0-9_-]{3,64}, which requires that the tenant identifier consists only of uppercase and lowercase letters, numbers, underscores (_), and hyphens (-), and has a length of at least 3 characters and no more than 64 characters. If the validation fails, the service will be rejected.

[0069] Before entering the business engine, callback events from enterprise collaboration platforms also record platform type, enterprise identifier, application identifier, member identifier or external contact identifier, session identifier, message identifier, receiving time, and processing status; duplicate events return existing processing results or idempotent success status, and do not create duplicate quotations, drawings, work orders, or dispatch tasks.

[0070] Tenant isolation can be achieved using independent directories, object storage namespaces, database schemas, table partitioning, or row-level tenant keys with mandatory access control. In one directory isolation implementation, the isolated storage directories are physically isolated by tenant identifier, with the storage path being data / tenants / . Each tenant has an independent subdirectory, the directory name of which is the tenant's tenant identifier. The price database directory stores the tenant's supplier quotation data, and the service provider database records the installation, after-sales, maintenance, inspection, and spare parts delivery resources that the current tenant is authorized to use. The product database directory stores the tenant's product specification data. The file database directory stores the tenant's business-generated result files (including bill of materials files, computer-aided design drawings, quotation files, and proposal files). The audit log directory stores each inference record of the tenant (including input questions, inference results, confidence levels, sources, and model versions).

[0071] Cross-tenant data access returns an explicit error and does not expose other tenants' data. In one implementation, tenant A's price library directory path is data / tenants / tenantA / supplier-library-v2 / libraries / , and tenant B's directory path is data / tenants / tenantB / . Tenant A's query request carries the tenant identifier tenantA, and the price library engine only reads data under the data / tenants / tenantA / directory. If tenant A's request attempts to access the data / tenants / tenantB / directory, the system returns an explicit error message "Insufficient permissions to access this tenant's data," and does not return any data content from tenant B.

[0072] In some implementations, the method further includes regional service routing based on geographic location context: receiving the current location authorized by the user, a manually entered project address, or an address identified from customer profiles or project files, standardizing it into latitude and longitude, administrative region, positioning accuracy, address source, and authorization status, and writing it into the geographic location context of the project parameter structure.

[0073] The service provider database records for each service resource the service provider's identifier, service type, coordinates or coverage area, service radius, types of LED displays that can be served, construction or after-sales qualifications, response time, historical ratings, pricing rules, availability, activation status, and tenant authorization status. The regional service routing engine calculates the distance between the project location and the service provider's location or coverage area, determines the service radius and capability qualifications, and sorts them according to distance, response time, qualifications, ratings, and tenant preferences.

[0074] For requests to install, provide after-sales service, repair, inspect, or deliver spare parts, the system returns candidate service resources using regional service matching cards. If no suitable candidate is found, the system returns a status indicating no available service and does not generate information about non-existent service providers. Location data is processed according to the principle of minimum necessity, and its use for new service matching ceases after authorization is withdrawn or a preset validity period expires.

[0075] like Figure 3 As shown, in one embodiment, a computer device is provided, which can be a server. The computer device includes a processor, memory, network interface, and database connected via a system bus; when the computer program is executed by the processor, it implements the functions or steps of the aforementioned dialogue agent interaction method for the LED display industry on the server side.

[0076] like Figure 4 As shown, in one embodiment, a computer device is provided, which can be a client. The computer device includes a processor, memory, network interface, display screen, and input device connected via a system bus; when the computer program is executed by the processor, it implements the functions or steps of the aforementioned dialogue agent interaction method for the LED display industry on the client side.

[0077] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0078] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0079] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0080] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A dialogue agent interaction method for the LED display industry, characterized in that, Includes the following steps: The system receives multi-source information input by the user within the dialog window of the graphical client. The multi-source information includes one or more of text, voice, images, file attachments, and user-authorized location information. It also verifies the login credentials, session token, or platform identity carried by the client and determines the tenant identifier and forms an authenticated tenant context on the server side. Normalization processing is performed on the multi-source information, and different types of multi-source information are uniformly extracted into a project parameter structure, and the location information is converted into the geographic location context in the project parameter structure; The semantic model is used to perform intent recognition on the project parameter structure, extract project parameter fields from the multi-source information, maintain the project parameter structure state machine including field source, confirmation status and version, and generate structured follow-up question cards in the dialog window to receive supplementary information from the user when the project parameter structure is missing key fields. Based on the intent recognition results, a target business capability is determined from multiple LED display industry-specific business capabilities. When the set of required fields corresponding to the target business capability in the project parameter structure meets the preset state access control, the corresponding business engine is invoked to perform business calculations. The progress and results of the business calculations are then fed back through status cards in the dialog window. Specifically, when the business intent is display industry knowledge Q&A, product selection consultation, fault diagnosis, or data acquisition, the industry knowledge retrieval engine is invoked based on the project parameter structure and the authenticated tenant context to form an evidence set from the display industry knowledge base, audited product specification library, and document library that the current tenant has the right to use. When the business intent involves engineering calculations, equipment compatibility, or configuration adaptation judgment, the corresponding deterministic business engine is further invoked, and a response or candidate data with source information is generated based on the evidence set and the return result of the deterministic business engine. When there is insufficient evidence, the candidate data fails to determine the unique target data according to the preset matching and selection rules, key adaptation parameters are missing, or access permission verification fails, the status is returned as pending supplementation, pending selection, pending confirmation, or undeliverable. When the business calculation result includes a bill of materials, the price library engine is called to match the price of each material. The price information of the materials that match the price is filled in, and the materials that do not match the price are marked as pending inquiry. All data access during the business calculation process carries the authenticated tenant context. When the user's intent involves installation, after-sales service, maintenance, inspection, spare parts delivery or regional service, the regional service routing engine is invoked based on the geographical location context in the project parameter structure to filter and sort candidate service resources from the service provider library. The storage space for business data, files, and audit records is isolated according to the certified tenant context, and the business calculation results and service matching results are archived to the isolated storage space of the current tenant. If the semantic model does not receive a successful execution result from the business engine, it will not output a message indicating that the business has been completed.

2. The method according to claim 1, characterized in that, The various LED display industry-specific business capabilities include bill of materials calculation capabilities, control system link planning and equipment selection capabilities, screen quotation capabilities, drawing generation capabilities, solution generation capabilities, inventory matching capabilities, and regional installation, after-sales or service routing capabilities. There is a real-time linkage between the bill of materials engine corresponding to the bill of materials calculation capability and the price database engine corresponding to the screen quotation capability; there is a predefined matching relationship between the regional service routing engine corresponding to the regional installation, after-sales or service routing capability and the geographical location context, service provider database, warehouse resource database or after-sales capability database; when the bill of materials calculation capability performs rounding up on the screen size according to the module specifications, the calculated actual screen size is written into the project parameter structure and marked as pending confirmation. After obtaining user confirmation, the actual screen size field satisfies the preset state access control. The control system selection engine corresponding to the control system link planning and equipment selection capability has a predefined linkage relationship with the project parameter structure, audited product specification library and bill of materials engine. The control system selection engine outputs the receiving card carrier sub-area, equipment category and quantity, port allocation, connection topology, cable list and items to be confirmed. The multiple LED display industry-specific business capabilities also include industry knowledge Q&A capabilities and controlled data delivery capabilities; the industry knowledge retrieval engine corresponding to the industry knowledge Q&A capability is used to retrieve evidence units based on product model, engineering parameters, question type and conversation context, and output one or more source information from document identifier, document version, source location, applicable product, validity status and access permissions; The knowledge objects of the display industry knowledge base include knowledge of one or more products or components such as LED display modules, light-emitting devices, driver chips, printed circuit boards, power supplies, enclosures, control system equipment, engineering structures, installation materials, software and configuration files, as well as one or more industry knowledge such as product naming, specifications, interface relationships, compatibility relationships, selection rules, manufacturing processes, installation and maintenance, fault diagnosis, safety specifications, industry standards, product versions and industry technology trends. The data matching engine corresponding to the controlled data delivery capability is used to perform product model, region, language, version and adaptation parameters matching for one or more of the following: product specifications, operation manuals, software installation packages and display configuration files. For the display configuration file, the adaptation parameters include at least one or more of the following: module pixels, scanning method, driving conditions, interface type, hub or receiver card model. When there is only one candidate file, or when multiple candidate files determine a unique target file through preset version priority, validity and adaptation rules, a deliverable file card is generated after the permission verification is passed. When multiple candidates exist and a unique target file cannot be determined according to the rules, a selection card is generated; if there is no match or no right to deliver, the file is not sent.

3. The method according to claim 1, characterized in that, Receiving multi-source information input by the user within the dialog window of the graphical client includes: Receive voice input and transcribe the voice input into text using an automatic speech recognition service; The system receives image input, extracts text from the image input using an optical character recognition module, and extracts scene visual features using a visual encoder or a multimodal visual model. Receive file attachment input, which supports multiple table formats; When the multi-source information comes from the enterprise collaboration platform, it receives messages or events transmitted from self-built applications, third-party applications, workbench web pages, customer contact tools or intelligent robots, and obtains at least one of the enterprise identifier, application identifier, member identifier or external contact identifier from the verified platform identity information. Receive the current location authorized by the user, the manually entered project address, or the project address identified from the project file, and normalize it into a geolocation context that includes one or more fields such as latitude and longitude, administrative region, positioning accuracy, address source, and authorization status.

4. The method according to claim 1, characterized in that, The project parameter structure includes project name, customer, installation location, screen width and height, usage environment, pixel pitch, module specifications, cabinet specifications, installation method, edging, back frame, steel material, structure type, profit margin, synchronous or asynchronous control method, input signal type and quantity, display screen partitions, full screen resolution, transmission distance, copper cable or fiber optic connection, multi-screen requirements, backup or redundancy requirements, field source, field status, and version field. The project parameter structure also includes a geographic location context, which includes one or more of the following: project address, latitude and longitude, administrative region, positioning accuracy, address source, and user authorization status.

5. The method according to claim 1, characterized in that, The invocation of the corresponding business engine to perform business calculations includes: When the estimated time for the business calculation is less than a preset first threshold, the result is returned synchronously; When the estimated time for the business calculation is greater than or equal to the preset first threshold and less than the preset second threshold, intermediate results are returned in blocks through streaming response. When the estimated time for the business calculation is greater than or equal to the preset second threshold, a background task is started and a task identifier is returned. Progress information is fed back to the dialog window through at least one of the following event channels: server event, WebSocket, chunked response, polling, message queue callback, or application message, interactive card, or smart robot callback of the enterprise collaboration platform. The result is called back after the calculation is completed.

6. The method according to claim 1, characterized in that, The status cards include various types such as plain text message cards, structured follow-up question cards, screen calculation progress cards, control system solution confirmation cards, drawing generation progress cards, solution document generation cards, inventory matching cards, price import result cards, and regional service matching cards. The structured follow-up question cards contain one or more sets of fields to be confirmed. The screen calculation progress cards include parameter completeness and screen preview information. The control system solution confirmation cards include candidate device links, receiving card carrier sub-regions and directions, port allocation, connection topology, cable quantity, redundancy method, source evidence, and items to be confirmed. The drawing generation progress cards include rendering progress and layer management information. The solution document generation cards include chapter options and customer information. The inventory matching cards include supplier, inventory quantity, price, and discount information. The price import result cards include pass / fail statistics, failure statistics, and error details, and are used to present file parsing, line-by-line verification, draft entry, and user activation results. The regional service matching cards include service provider name, service type, distance, service radius, response time, qualification level, rating, availability status, and executable operations. The status cards are converted into native graphical client cards, enterprise collaboration platform application messages, interactive cards, or intelligent robot replies based on the client's capabilities, while maintaining the correspondence between task identifiers, result statuses, data sources, and executable operations; The status card also includes a knowledge response card, a source evidence card, a candidate data selection card, and a document delivery card. The knowledge response card marks the search facts, business engine calculation results, and information to be confirmed respectively. The source evidence card records the document name, version, page number or content location, and validity status. The document delivery card records the document identifier, version, matching basis, access permissions, and applicable conditions.

7. The method according to claim 1, characterized in that, The regional service routing engine performs the following steps: It obtains the user-authorized current location, the manually entered project address, or the project address identified from the project file, and converts it into the geographic location context; it queries the service provider database, recording service type, coverage area, service radius, LED display type capabilities, construction or after-sales qualifications, response time, rating, pricing rules, availability status, and tenant authorization status; it calculates the distance between the project location and the service provider location or coverage area and determines the service radius, sorting candidate service providers according to distance, response time, qualifications, rating, and tenant preferences; if no suitable candidate service provider is found, it outputs a "no available service" status and does not generate information about non-existent service providers.

8. The method according to claim 1, characterized in that, It also includes real-state access control steps: When no quote is matched in the semantic model, the status of pending inquiry is output without fabricating supplier, price, inventory, discount or delivery information; When the business engine is unavailable, output a message indicating that the service is temporarily unavailable while retaining the user's request; When the regional service routing engine does not match a candidate service provider that meets the service radius, capability, qualification or tenant authorization conditions, it outputs a state of no available service and does not fabricate service provider, distance, qualification, price or reservation status. When the industry knowledge retrieval fails to obtain an evidence set that meets the source validity requirements, the content generated by the semantic model will not be marked as a verified response. When the candidate data fails to determine a unique target data according to the preset matching and selection rules, the configuration adaptation parameters are not verified, the source of the file is untraceable, the file has expired, or the current tenant does not have delivery permissions, the corresponding file will not be sent, and the status of pending selection, pending supplementation, pending source supplementation, expired, or undeliverable will be returned.

9. The method according to claim 1, characterized in that, The tenant identifier is determined by the server based on login credentials, session tokens, or platform identity identifiers, and format and authorization relationship verification is performed. When the platform identity comes from an enterprise collaboration platform, the server verifies the authorization relationship between the enterprise identity and the application identity, and maps the member identity or external contact identity to the user identity, role permissions and customer identity under the current tenant; Business data, files, and audit records are logically or physically isolated according to tenant identifiers. The isolation includes at least one of the following: independent directories, object storage namespaces, database schemas, table partitions, or row-level tenant keys with mandatory access control. Bind the price database, product database, service provider database, file database, and audit logs to tenant identifiers, tenant storage namespaces, and access control policies; Cross-tenant data access returns an explicit error and does not expose other tenants' data.

10. The method according to claim 1, characterized in that, The industry knowledge retrieval engine and the data matching engine process user requests according to the following steps: The system performs segmentation, fieldization, and versioning on display industry documents, product specification data, engineering rules, and file metadata related to LED display modules, light-emitting devices, driver chips, printed circuit boards, power supplies, enclosures, control system equipment, engineering structures, installation and maintenance, or industry technical information to form searchable evidence units. Based on the project parameter structure, it generates one or more search conditions from keyword conditions, semantic conditions, and structured field conditions to obtain candidate evidence units. These candidate evidence units are then rearranged and filtered according to source level, version validity, product applicability, field consistency, and tenant access permissions. When a user request is for knowledge-based questions, the filtered evidence units are bound to the calculation results returned by the deterministic business engine to generate a response, and the evidence units referenced in the response are recorded. When a user request is for data retrieval, candidate files are validated for product model, region, language, version, and compatibility parameters. If there is only one candidate file, or multiple candidate files are determined to be a unique target file based on preset version priority, validity, and compatibility rules, the unique target file is sent under the condition that the current tenant has the right to deliver it, and the file version, source, matching basis, and session identifier are written into the audit log.

11. A dialogue-based intelligent agent interaction system for the LED display industry, characterized in that, It includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 10.

Images