Cooperative BOM data analysis and verification system and method based on AI model

Abstract

The invention relates to the technical field of BOM data analysis, and discloses a collaborative BOM data analysis and verification system and method based on an AI model.The collaborative BOM data analysis and verification system based on the AI model comprises an intelligent document sensing and structure analysis module, the intelligent document sensing and structure analysis module monitors and identifies a list file, and sends the list file to a server; and the AI assistant dialogue engine module adopts a language model to interact with an engineer to analyze the document in the unified format field by field. According to the method, a field-level approval mechanism is introduced, each suggestion can be modified after being confirmed by a user, and the user can execute operations such as rejection, remark and pending instead of rechecking only in the final stage, so that the sovereignty, the safety and the traceability of the user are improved, the liability boundary of AI aided design is cleared, the AI is evolved from a tool to an engineering partner with learning ability, and the work efficiency is improved. Interaction intelligent experience is obviously enhanced, and adaptability is high.

Description

Technical Field

[0001] The present invention relates to the technical field of BOM data analysis and verification, and specifically to a collaborative BOM data analysis and verification system and method based on an AI model. Background Art

[0002] Bill of Materials (BOM) is a technical document that describes the composition of a company's products. In the capital-processing industry, it shows the structural relationship between the final assembly, sub-assembly, components, parts, and even raw materials of the product, as well as the required quantity. In the chemical, pharmaceutical and food industries, the product composition describes the main raw materials, intermediates, auxiliary materials, their formulas and the required quantities. BOM expresses the product composition represented by a diagram in the form of a data table. It is an important control document in the MRP calculation process of the MRPII system.

[0003] During the development of electronic products, design engineers need to compile both single-board bills of materials (BOMs) and product hardware lists for subsequent procurement, manufacturing, assembly, and testing. However, due to the wide variety and large quantity of hardware BOMs, inconsistent formats, and the error-prone nature of manual entry, various BOM issues often arise. Traditional manual verification is inefficient and error-prone, hindering rapid batch review. Existing tools generally have fixed logic and cannot handle complex exceptions. Fuzzy semantics solutions based on AI-based large language models (such as CN119004372A) are mostly limited to automated analysis, lacking key design features such as interaction mechanisms with engineers, modification approval processes, and user preference memory, making it difficult to cover the entire engineering decision-making process. Summary of the Invention

[0004] The present invention provides a collaborative BOM data analysis and verification system and method based on an AI model, which has the beneficial effect of being able to interact with engineers and cover the entire process of engineering decision-making. It solves the problems mentioned in the above background technology that are not conducive to rapid batch review, remain at the automatic analysis level, lack key designs such as interaction mechanisms with engineers, modification approval flows, and user preference memory mechanisms, and are difficult to cover the entire process of engineering decision-making.

[0005] The present invention provides the following technical solution: a collaborative BOM data analysis and verification system based on an AI model, comprising:

[0006] An intelligent document perception and structure analysis module, which monitors and identifies the manifest file and converts the output into a document in a unified format;

[0007] An AI assistant dialogue engine module, which uses a language model to interact with engineers to analyze documents in a unified format field by field, thereby assisting engineers in understanding the contents of the manifest file;

[0008] A semantic classification and field specification module, which marks semantically ambiguous fields in a unified format document and triggers the AI assistant dialogue engine module to issue a suggestion message for the engineer to confirm;

[0009] An intelligent verification and abnormality diagnosis module, which interfaces with fields in a device database, verifies and diagnoses a unified format document based on the fields in the device database, and generates corresponding suggestion records for engineers to process;

[0010] A deployment and resource scheduling optimization module embeds and integrates the AI assistant dialogue engine module and the language model, and the embedded integrated AI assistant dialogue engine module analyzes unified format documents in an offline and low-power environment.

[0011] As an optional solution of the collaborative BOM data analysis and verification system based on the AI model of the present invention, wherein: the AI assistant dialogue engine module includes a user interaction layer, a semantic intermediate processing layer and a reasoning and response layer;

[0012] The user interaction layer uses PySide2 to provide users with a natural language interaction window;

[0013] The semantic intermediate processing layer parses the unified format document through the session state manager, prompt builder and instruction intent recognizer;

[0014] The reasoning and response layer uses the model service interface and the streaming output manager to dynamically add tags to the parsed unified format documents in real time.

[0015] As an optional solution to the collaborative BOM data analysis and verification system based on the AI model of the present invention, the model service interface uses the mistral.cpp framework to load the language model into the local computer in GGUF format, and uses INT4 quantization to optimize inference efficiency.

[0016] The model service interface automatically switches and hot loads language models based on the context or task type of the unified format document.

[0017] As an optional solution of the collaborative BOM data analysis and verification system based on the AI model of the present invention, the streaming output manager pushes the level-by-level token results returned by the language model to the AI assistant dialogue engine module through SSE based on the Streaming Token inference mechanism;

[0018] The AI assistant dialogue engine module dynamically adds tags in real time based on the level-by-level token results.

[0019] As an optional solution of the collaborative BOM data analysis and verification system based on the AI model of the present invention, wherein: the deployment and resource scheduling optimization module includes a model operation interface layer and an inference scheduling and caching layer;

[0020] The model operation interface layer includes using a model loader to load the GGUF format weight file of the local language model into the memory, using a memory mapping mechanism to achieve efficient access and on-demand retrieval of the weight file, and building the local inference engine compiled from the language model into a static link library and embedding it into the desktop application front-end program;

[0021] The inference scheduling and caching layer includes a thread pool scheduling mechanism, which encapsulates each inference request as an asynchronous task object and submits it to the thread pool managed by ThreadPoolExecutor for concurrent execution. After all requests are queued, they are scheduled and processed according to task priority.

[0022] As an optional solution of the collaborative BOM data analysis and verification system based on the AI model of the present invention, wherein: the deployment and resource scheduling optimization module also includes a model management and upgrade subsystem layer and a local log and authority control subsystem layer;

[0023] The model management and upgrade subsystem layer includes automatically searching for language model files that meet the format through the specified model folder when the user switches multiple language models. When dynamically replacing the currently loaded language model, it unloads the current model weights and clears the related memory mapping through the resource release interface, then reloads the new GGUF model file, and refreshes the inference context cache and user session records.

[0024] The local log and authority control subsystem layer is used to write each round of conversation records into the local database.

[0025] As an optional solution of the collaborative BOM data analysis and verification system based on the AI model of the present invention, it also includes an approval mechanism and status tracking module, which establishes an auxiliary decision-making mechanism based on combining the suggestion information issued by the AI assistant dialogue engine module with the user's engineering behavior;

[0026] The approval mechanism and status tracking module includes a suggested status data structure submodule, an approval interaction interface submodule, and a status mark visualization submodule;

[0027] The suggestion status data structure submodule records and controls the entire life cycle of AI-generated suggestions, records all changes in the local log database, and forms a history of suggestion status changes.

[0028] The approval interaction interface submodule provides a button operation or dialogue instruction triggering mechanism, allowing users to trigger state change behavior by clicking a button or using natural language instructions;

[0029] The status mark visualization submodule is used to intuitively present the current processing status of each suggestion in the human-computer interaction interface.

[0030] As an optional solution of the collaborative BOM data analysis and verification system based on the AI model of the present invention, it further includes a memory mechanism and a personalized adaptation module, which are used to train the AI assistant in the AI assistant dialogue engine module;

[0031] The memory mechanism and personalized adaptation module are used to track and record the user's feedback behavior on the system-generated suggestions, form a data basis for preference learning and behavior modeling, automatically inject user preferences and historical context into the large model prompt words, sort the results, and optimize the sorting of multiple suggestions or alternative devices or instructions returned by the large model, and build a long-term preference modeling and behavior memory storage for users, so that the AI assistant in the training AI assistant dialogue engine module has intelligent behavior of memory.

[0032] As an optional solution of the collaborative BOM data analysis and verification system based on the AI model of the present invention, it further includes a multi-window parallel view module, wherein the multi-window parallel view module is provided with a main table view, a dialogue window, an approval window and a graphics window;

[0033] The data in the main table view, dialogue window, approval window and graphic window are linked to each other, and clicking on a field automatically links other windows to update.

[0034] The processing method of the collaborative BOM data analysis and verification system based on the AI model includes the following steps:

[0035] S1: When a hardware BOM file is opened, the AI assistant window is automatically activated and the parsed file is read to initialize the user session. The user's memory configuration field preferences, operation habits, and project context are read to generate a unified data structure for the interactive module to call.

[0036] S2: Marking of documents through three triggering methods: user questions, field clicks, and automatic checks;

[0037] S3: Generate model inputs through the Prompt Builder and dynamically construct customized prompts based on the combined field content, user preferences, and project memory;

[0038] S4: local large language model operation and adjusted model parameters operation;

[0039] S5: The AI assistant generates answers including field explanations, verification judgments, modification suggestions, and recommended alternative devices. Responses are streamed out in the form of Streaming Tokens, allowing users to interrupt generation and re-ask.

[0040] S6: The user approval mechanism is triggered, and the "confirm / reject" operation is executed through buttons or dialogues. The user operation triggers the behavior collection module to record the approval result. The approval result is marked in the hardware BOM file and written into the audit log. The user's operation behavior and preference trends are recorded to extract long-term structural preferences.

[0041] S7: Subsequent conversations refer to the memory for semantic optimization and response sorting. The response order is optimized based on historical behavior, user profiles are constructed, and long-term collaborative efficiency optimization is achieved.

[0042] S8: Application export approval log, personalized preference log, result file model record file.

[0043] The present invention has the following beneficial effects:

[0044] 1. This AI-based collaborative BOM data analysis and verification system and method introduces a field-level approval mechanism. Each suggestion must be confirmed by the user before it can be modified. Users can perform operations such as rejection, remarks, and pending, rather than just reviewing in the final stage. This increases user sovereignty, security, and traceability, clarifies the boundaries of responsibility for AI-assisted design, and evolves AI from a tool to a learning engineering partner, significantly enhancing the interactive intelligent experience and making it highly adaptable.

[0045] 2. This AI-based collaborative BOM data analysis and verification system and method supports local offline operation and can be run on the enterprise intranet or local terminals to ensure data security. Compared with existing systems that rely more on cloud API calls and have the risk of data leakage, it solves the problem of enterprise implementation of AI in BOM processing, has industry deployability, and is particularly suitable for confidential R&D environments. It creates a model for engineering-level system design and is different from general-purpose natural language analysis tools.

[0046] 3. This AI-based collaborative BOM data analysis and verification system and method uses a front-end embedded assistant architecture with independent workflow and operation status. It breaks through traditional command-based AI interaction and actively senses user-opened files in assistant mode and follows up in real time, simulating the conversational review behavior of real engineers. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flow chart of the AI-collaborative BOM data analysis and auxiliary verification system.

[0048] Figure 2This is a flowchart of the AI assistant dialogue engine module.

[0049] Figure 3 This is a flowchart of the local deployment and resource scheduling optimization module.

[0050] Figure 4 This is a flowchart of the approval mechanism and status tracking module.

[0051] Figure 5 Schematic diagram of the memory mechanism and personalized adaptation module.

[0052] Figure 6 The figure is a flowchart of the overall collaborative BOM data analysis and auxiliary verification method based on the AI model. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0054] Example 1

[0055] See also Figures 1-6 , wherein the collaborative BOM data analysis and verification system and method based on the AI model includes:

[0056] Intelligent document perception and structure analysis module: The intelligent document perception and structure analysis module monitors and identifies the list file and outputs it into a unified format document;

[0057] Monitors BOM and hardware list files. When a user opens a BOM or hardware list file, it automatically identifies the file (based on field keywords, format characteristics, etc.). It uses openpyxl or pandas to parse the file content, supports merging cells, recovering the structure of deformed tables, and outputs a standard two-dimensional data structure. The output format is unified into a DataFrame structure for downstream modules to call.

[0058] It should be noted that pandas is a tool based on NumPy, which was created to solve data analysis tasks. Openpyxl is a Python library that focuses on processing modern Excel files and is mainly used to automate Excel documents.

[0059] The AI assistant dialogue engine module uses a language model to interact with engineers and analyze documents in a unified format field by field, thereby helping engineers understand the contents of the manifest file;

[0060] The AI assistant dialogue engine module is embedded in the front-end dialogue window, supporting mixed Chinese and English questions, field clicks to trigger contextual descriptions, and fuzzy query commands, such as "What's unusual about this line?"

[0061] It should be noted that the language model is the open-source Mistral7B model, which is deployed on a local PC and runs offline. The dialogue interaction process is implemented based on a contextual memory buffer mechanism, supporting multiple rounds of semantic interaction and continuous state tracking.

[0062] Specifically, the AI assistant dialogue engine module includes a user interaction layer, a semantic intermediate processing layer, and a reasoning and response layer;

[0063] The user interaction layer uses PySide2 to provide users with a natural language interaction window;

[0064] The semantic intermediate processing layer parses the unified format document through the session state manager, prompt builder and instruction intent recognizer;

[0065] The reasoning and response layer uses the model service interface and streaming output manager to dynamically add tags to the parsed unified format documents in real time;

[0066] The model service interface uses the mistral.cpp framework to load the language model locally in GGUF format and uses INT4 quantization to optimize inference efficiency.

[0067] The model service interface automatically switches and hot-loads language models based on the context or task type of unified format documents;

[0068] The streaming output manager pushes the level-by-level token results returned by the language model to the AI assistant dialogue engine module through SSE based on the Streaming Token inference mechanism;

[0069] SSE, short for Server-Sent Events, is a one-way communication mechanism based on HTTP that allows a server to proactively push real-time data streams to a client (typically a web browser). SSE only supports one-way data push from the server to the client.

[0070] The AI assistant conversation engine module dynamically adds tags in real time based on the level-by-level token results;

[0071] Among them, the AI assistant dialogue engine module is responsible for the core subsystem of human-computer natural language interaction. Its goal is to build a floating window intelligent assistant embedded in the desktop to assist engineers in performing interactive operations such as field interpretation, abnormal diagnosis, and recommended approval through dialogue when processing BOM and hardware list files. Each submodule is decoupled from the event queue mechanism using asynchronous calls. The three layers of UI interaction, semantic processing, and model reasoning can be independently tested and replaced, and have good engineering modularity. It realizes the systematic integration of engineering of local offline semantic assistants, and has the closed-loop capability of semantically driven behavior, dialogue memory state mechanism, and engineering linkage operation, including the following functions:

[0072] Field explanation: When the user clicks on a form field, the assistant automatically explains its meaning and typical format;

[0073] Anomaly diagnosis: When a user asks "Is there a problem with this line?", the assistant analyzes the database and diagnoses the anomaly.

[0074] Component replacement recommendation: Enter "replace a suitable capacitor" and the assistant will query the database and return suggestions;

[0075] Operation command trigger: The user enters "Confirm suggestion" and the assistant drives the approval process update;

[0076] Multi-round memory question and answer: continuous contextual questions without the need to repeatedly input device information;

[0077] Session log tracking: All questions, answers and operations are automatically recorded and can be traced back and exported;

[0078] Linked interaction: When the user clicks a field in the main table or selects a device row, the assistant window can automatically insert information;

[0079] The AI assistant dialogue engine module includes a user interaction layer (UI interface), a semantic intermediate processing layer, and a reasoning and response layer;

[0080] The user interaction layer (UI interface) provides users with a natural language interaction window, mainly including:

[0081] Use PySide2 to implement a floating assistant interface;

[0082] The interface includes input box, send button, scroll response area, linkage interaction area, etc.

[0083] The response area supports Markdown rendering (bold / hyperlink / code block);

[0084] When the user clicks on the main BOM table field, the assistant is automatically passed in as context injection;

[0085] Among them, the semantic intermediate processing layer includes the session state manager, prompt builder and instruction intention recognizer;

[0086] Session State Manager: Maintains the session state of the current BOM document, generates a separate session_id when each document is opened, uses local .json / SQLite files to save logs, and provides session loading, clearing, and export capabilities;

[0087] Prompt Builder: Generates standardized prompts for model understanding, automatically identifies field semantics, and constructs templates. For example, it dynamically generates package, voltage, and description fields, transforming structured data into natural language input. For example, "The current device is: capacitor, 0603 package, 10uF / 16V, please determine whether it meets the requirements." By incorporating dynamic prompt selection after user preference memorization and field naming deviations as contextual prompts, the final input is fed into the Mistral7B model.

[0088] Command intent recognizer: Determines whether user input is a specific system command, such as approval, recommendation, or verification. It uses MiniLM or BERT to vectorize the input sentence and perform Top-K similarity matching with a command corpus ("confirm," "change," and "replace recommended capacitor"). The command corpus supports dynamic learning and user customization, and supports embedding field context into command vector calculations, such as "click the capacitor value field and say 'change'."

[0089] The inference and response layer includes the model service interface and the streaming output manager;

[0090] Model service interface: Use the mistral.cpp framework to load the Mistral7B model locally in GGUF format, using INT4 quantization to optimize inference efficiency. This interface supports automatic model switching and hot loading based on user context or task type.

[0091] Streaming Output Manager: The streaming output manager is implemented based on the StreamingToken inference mechanism. It pushes the token-by-token results returned by the language model to the front-end assistant interface through SSE. The interface uses the QTextBrowser control for batch rendering and real-time updates, supports user interaction interruption (such as the "Stop Generation" button), and dynamically inserts field highlights, text underlines, jumpable hyperlinks, and other enhanced visual feedback during the display process to improve human-computer interaction efficiency.

[0092] The semantic classification and field specification module marks semantically ambiguous fields in documents in a unified format and triggers the AI assistant dialogue engine module to issue suggestions for engineers to confirm.

[0093] The semantic classification and field specification module uses a rule engine and regular expressions to identify common field aliases, such as "PN / PartNo. / Number → Material Code". When the semantics of a field are ambiguous, the AI assistant is triggered to pop up a confirmation suggestion, such as "Is the Value field a resistor value or a device description?" The semantic classification and field specification module supports the automatic construction of a "field semantic mapping dictionary" for reuse by the verification and export modules. Unstructured fields are vectorized and encoded for subsequent searches using the BGE-m3 model. The semantic classification and field specification module is used to explicitly mark primary key fields, polysemous fields, and ambiguous fields in documents in a unified format;

[0094] The intelligent verification and abnormality diagnosis module connects to the fields in the device database, verifies and diagnoses the unified format document based on the fields in the device database, and generates corresponding suggestion records for engineers to process;

[0095] The intelligent verification and anomaly diagnosis module connects to the internal device database (MySQL / SQLite) to perform field validity matching. It uses a rule base to check for duplicate bit numbers, illegal formats, recommended voltages, tolerances, and packages, and identifies redundant part numbers. It uses a large model to analyze context and provide "natural language suggestions." All anomalies are automatically recorded as suggestions, assigned a suggestion ID (such as SUG-00001), and pushed to the approval module. Diagnosis results include status tags and a manual intervention interface. Each suggestion includes a number, source, and contextual prompt information.

[0096] Approval mechanism and status tracking module: Each suggestion item is accompanied by a status label, including pending confirmation, approved, rejected, and skipped. Users can click to confirm, reject, and note the reason for rejection. All operations are recorded in real time in the "Approval Log", including the suggestion ID, original value, new value, timestamp, user, and decision status. An approval UI is provided, which can filter all "pending approval items" for batch processing;

[0097] The deployment and resource scheduling optimization module integrates the AI assistant dialogue engine module with the language model. The embedded integrated AI assistant dialogue engine module analyzes documents in a unified format in an offline and low-power environment.

[0098] The deployment and resource scheduling optimization module includes the model operation interface layer and the inference scheduling and caching layer;

[0099] The model operation interface layer uses a model loader to load the GGUF format weight file of the local language model into memory, uses a memory mapping mechanism to achieve efficient access and on-demand retrieval of the weight file, and builds the local inference engine compiled from the language model into a static link library and embeds it into the desktop application front-end program;

[0100] The inference scheduling and caching layer uses a thread pool scheduling mechanism to encapsulate each inference request as an asynchronous task object and submit it to the thread pool managed by ThreadPoolExecutor for concurrent execution. After all requests are queued, they are scheduled and processed according to task priority.

[0101] The deployment and resource scheduling optimization module also includes the model management and upgrade subsystem layer and the local log and permission control subsystem layer;

[0102] The deployment and resource scheduling optimization module is a key subsystem responsible for the semantic interaction capabilities of large language models that operate fully localized and offline. Its goal is to enable AI assistants to deploy models locally without calling any external APIs. This ensures that the model is tightly integrated with the system front-end, enabling immediate use without data leaving the local machine. It retains interfaces for hot model replacement and upgrades, supports subsequent model switching, and supports performance expansion. Specifically, it includes the following features:

[0103] Local deployment capability: Supports embedding and integration of models with front-end assistant processes, allowing large language models to be run offline on personal PCs, ensuring data privacy and local computing needs.

[0104] The inference process is embedded in the front-end: Model calls are directly executed by the helper program, without the need for REST APIs, servers, or background service processes, improving deployment simplicity and security.

[0105] Model management and upgrade: Supports the coexistence of multiple model files, and enables hot loading and switching through configuration or interface, facilitating subsequent expansion of model capabilities (such as upgrading from 7B to 13B);

[0106] Streaming inference control: The Streaming Token push mechanism is used during the inference process, generating and outputting data simultaneously, reducing waiting time and improving interaction fluency.

[0107] Cache mechanism support: Provides local cache and fast return mechanism for repetitive prompts, common device explanations, etc., to avoid repeated model calls and reduce resource waste;

[0108] Data closure and permission control: Inference, cache, and log data are all stored locally, independent of external networks. User operation permissions and sensitive log access control policies can be set.

[0109] The model management and upgrade subsystem layer includes automatically searching for language model files that match the format through the specified model folder when the user switches between multiple language models. When dynamically replacing the currently loaded language model, it unloads the current model weights and clears the related memory mapping through the resource release interface, then reloads the new GGUF model file and refreshes the inference context cache and user session records.

[0110] The local log and permission control subsystem layer is used to write each round of conversation records into the local database;

[0111] Specifically, the model management and upgrade subsystem layer calls the model loader upon system startup to load the local Mistral7B model's GGUF-format weight file into memory. Memory mapping mechanisms, such as mmap(), enable efficient access and on-demand retrieval of the weight file. This approach enables local preparation and initialization of large language models on the user terminal without the need for an external server or REST interface, significantly improving inference response efficiency and ensuring data remains locally within the user. The local inference module supports embedded integration deployment. This involves building the local inference engine compiled from the Mistral7B model into a static library (such as a .a or .lib file) and embedding it in the desktop application front-end for direct call, bypassing traditional REST API or server service processes. This improves runtime response speed and overall system architecture compactness. The inference interface encapsulation shields the complexity of the underlying model inference process through standardized calling methods, encapsulating details such as multi-parameters, asynchronous execution, and resource scheduling into a callable function interface for the front-end. The front-end program simply passes in user input and model parameters to complete inference tasks, enhancing system modularity and callability, facilitating integration into desktop assistant interfaces or scripted processes. An interface encapsulation mechanism supports automatic splicing of prompts (fields + user preferences) for structured field input. Parameter control supports flexible configuration of multiple core parameters of the language model generation strategy, including generation length (max_tokens), randomness control coefficient (temperature), sampling coverage parameter (top_p), etc., which can be dynamically set when called through the front-end interface or user configuration file. This mechanism avoids hard-coding parameters and allows dynamic selection of generation strategies based on task context, user preferences, or predefined templates, improving system controllability and adaptability. It is used to build a trusted local operating environment, shielding the ability to call external models and ensuring that all inference behavior is derived solely from local input.

[0112] Specifically, the inference scheduling and caching layer includes asynchronous task scheduling, suitable for local inference deployment scenarios on desktops. The system utilizes a thread pool scheduling mechanism, encapsulating each inference request as an asynchronous task object and submitting it to a thread pool managed by ThreadPoolExecutor for concurrent execution, thus avoiding blocking the main thread. The task scheduling system dynamically adjusts thread priority based on inference task type. Scheduling optimization for "conversational requests > batch generation requests" can be combined with a caching mechanism to implement "cache-driven task preemptive scheduling." The interrupt control mechanism for streaming token output enables linkage between asynchronous scheduling and streaming output (with interruption capabilities if tasks are not completed), improving overall system responsiveness and resource utilization. This is suitable for scenarios where multiple windows and multiple sessions concurrently call local language models. A thread-safe task queue with a first-in-first-out (FIFO) structure manages all inference requests. After enqueuing, all requests are scheduled based on task priority. The system distinguishes between critical and non-critical tasks. For example, a request for explanation triggered by a field click will have a higher scheduling priority than a user's natural language question. This mechanism ensures rapid response to high-priority tasks, enhancing the user interaction experience and concurrent system task processing efficiency. By performing a hash digest operation on the prompt input by the user, a corresponding unique identifier (Key) is generated and quickly matched in the local cache database. If the cache matches, the previously generated response result is directly returned, skipping the inference process. If it misses, regular inference is performed and the new result is written to the cache. This mechanism effectively reduces repeated calculations, improves response efficiency, and reduces system resource usage.

[0113] Specifically, the model management and upgrade subsystem supports users quickly switching between multiple model versions. It monitors a designated model folder, such as the "models / " directory, automatically searches for all model files that match the required format, and dynamically generates a list of available models for display on the front-end. This allows users to dynamically replace the currently loaded language model during runtime, preventing context carryover and inference state conflicts. During the model switching process, the resource release interface is first called to unload the current model weights and clear the associated memory mapping. The new GGUF model file is then reloaded, and the inference context cache and user session history are refreshed, ensuring operational isolation and a clean environment for multi-model switching. This process integrates with the user interface to provide a progress bar and switch status prompts, enhancing operational controllability and a visual experience. It tracks the call history of language models during runtime, supporting performance comparison and analysis of multiple model versions, operational statistics, and behavior tracing. Each model call automatically records the call time, model name, version number, generated task type, and call result, and writes the data to a local log database, such as SQLite. This module assists in subsequent model switching strategy optimization, user behavior analysis, and call performance statistics, improving maintainability and intelligence. To prevent misoperations, security restrictions are added. Only administrators can switch or import new models (authorization password required). This ensures that AI services run securely within a trusted scope and locally.

[0114] Specifically, local logs and permission controls include writing each round of conversation records (prompt, response (response is the answer generated by the AI model based on the prompt), time, model name, and timestamp) into the local database for subsequent tracing and analysis output. Log export capabilities are designed for compliance, and administrators can export audit reports in .csv / .md / .json formats for security audits or version comparisons. Automatic networking and remote model downloads are prohibited, and the entire operation is closed locally (no public network sockets / requests). A "clean log" function is provided to regularly delete historical conversation records from 30 days ago, and support filtering and clearing by keywords / fields / users to save space.

[0115] It also includes an approval mechanism and status tracking module, which establishes an auxiliary decision-making mechanism based on combining the suggestion information issued by the AI assistant dialogue engine module with user engineering behavior;

[0116] The approval mechanism and status tracking module includes the proposed status data structure submodule, the approval interaction interface submodule, and the status mark visualization submodule;

[0117] The suggestion status data structure submodule records and controls the entire life cycle of AI-generated suggestions, and records all changes in the local log database to form a history of suggestion status changes.

[0118] The approval interaction interface submodule provides a button operation or dialogue instruction trigger mechanism, allowing users to trigger state change behavior by clicking a button or using natural language instructions;

[0119] The status mark visualization submodule is used to intuitively present the current processing status of each suggestion in the human-computer interaction interface.

[0120] The approval mechanism and status tracking module is used to effectively combine AI-generated suggestions with user engineering behaviors. This establishes an intelligent decision-making support mechanism that combines "structured status tracking, manual confirmation, and operation logging." This gives users the "right to approve" AI output results, ensuring process compliance, security, and traceability. Specifically, it includes the following functions:

[0121] Suggestion status management: Each suggestion is bound to a unique status: pending confirmation, confirmed, rejected, skipped, and withdrawn;

[0122] User approval interaction: Users perform approval actions on suggestions through dialogue instructions or button operations;

[0123] Status visual marking: Mark the "Confirmed / Rejected" field in the main table (such as color, icon);

[0124] Approval log tracking: automatically record approval behavior, operator, time, suggestion ID, decision results and other information;

[0125] Batch approval function: supports one-click confirmation / rejection of similar suggestions to improve efficiency;

[0126] Approval result export: Approval records can be exported to files for archiving or external review;

[0127] Specifically, the suggestion status data structure submodule is used to record and control the status changes of AI-generated suggestions throughout their lifecycle, ensuring that all suggestions are controllable, traceable, and revocable. The system assigns a unique identifier to each suggestion (e.g., SUG-0001) and provides a unified status update interface that accepts the suggestion ID, new status type, and user identity as parameters to complete the suggestion status update process. All changes are recorded in the local log database, forming a suggestion status history that can be used for approval backtracking, version comparison, and user behavior analysis.

[0128] Specifically, the Approval Interaction Interface submodule provides a button-based or dialogue-based triggering mechanism, allowing users to trigger state changes through button clicks or natural language commands. Graphically, each suggestion displays "Confirm / Reject / Skip" buttons to the right. Conversationally, it supports semantic commands such as "Confirm suggestion 2," "Reject alternative suggestions," and "Approve all encapsulated modification suggestions." The intent recognition module then converts the natural language into standard commands. Upon successful operation, the assistant returns feedback: "The modification suggestion has been confirmed, and the system has updated the status."

[0129] The Status Marker visualization submodule visually displays the current status of each suggestion within the user interface. The system embeds suggestion status into the corresponding cell of the main table interface using color coding (e.g., green for confirmed, red for rejected, and yellow for pending) and icon prompts. Hovering the mouse over a suggestion field displays a pop-up notification containing the time the status changed and the user who performed the action. When a suggestion status is updated, the system automatically refreshes the visual marker area, eliminating the need for page reloads or manual intervention.

[0130] The Approval Log Recording submodule provides an interface for audit writing, exporting, and backtracking, recording the complete operational trajectory of each approval action and supporting subsequent review, export, and reuse. By storing approval records in SQLite and supporting single-item export and conditional export (e.g., exporting all "rejected" suggestions), approval operation records are only added, not modified, ensuring data auditability.

[0131] It also includes a memory mechanism and a personalized adaptation module, which are used to train the AI assistant in the AI assistant dialogue engine module;

[0132] The memory mechanism and personalized adaptation module are used to track and record users' feedback on system-generated suggestions, forming a data foundation for preference learning and behavior modeling, automatically injecting user preferences and historical context into the large model prompt words, sorting the results, and optimizing the sorting of multiple suggestions or alternative devices or instructions returned by the large model, and building a long-term preference modeling and behavior memory storage for users, so that the AI assistant in the training AI assistant dialogue engine module can have intelligent behavior of memory.

[0133] The memory mechanism and personalized adaptation module elevate the AI assistant from a "general-purpose large model" to an "engineering assistant familiar with user habits," enabling true intelligent human-machine collaboration. Its goal is to gradually establish a "user profile" and a "semantic preference database" by recording user behavior and identifying usage preferences. It supports hierarchical management of project-level and user-level memories. All memory content is stored locally and can be transparently viewed or cleared. Specifically, it includes the following functions:

[0134] User behavior memory: records the user's actions on suggestions (confirmation, rejection, modification, etc.);

[0135] Preference learning: Automatically learn preferences (such as packaging priority and brand preference) based on user historical behavior;

[0136] Prompt adaptation: automatically embed user habits and contextual features when constructing prompt words;

[0137] Response sorting optimization: personalized sorting and screening of recommendation results;

[0138] Multi-dimensional memory structure: supports the establishment of independent memory cache based on the dimensions of "user / project / device";

[0139] Visual management: provides functions for viewing, editing, and clearing memory items to ensure controllability;

[0140] Specifically, the user behavior collection submodule is used to track and record user feedback on system-generated suggestions, forming a data foundation for preference learning and behavior modeling. Each suggested user action (such as confirmation, rejection, modification, and ignoring) will be converted into a structured behavior log item, containing information such as the suggestion number, the operating user, the operation time, the suggested field, and the final accepted value, and will be categorized and written into the local user behavior log database. The system supports triggering log collection through front-end interaction (such as clicking the confirmation button) or parsing the assistant's natural language interaction instructions, building a unified behavior collection channel. This module can provide behavioral data support for subsequent suggestion sorting optimization, user preference customization, and system recommendation strategies.

[0141] Specifically, the personalized prompt builder automatically injects user preferences and historical context into large model prompts to improve inference accuracy and construct templates after personalization. For example, if you are a BOM assistant familiar with the preferences of user "XXX", who prefers domestic components, 1% resistors, and prefers brand XXX, and the current component is a 10uF capacitor of brand XXX, please determine whether it is recommended and whether it should be replaced with a domestic component. All personalized injection logic is completed by the prompt splicer, which embeds and generates a summary of historical preferences.

[0142] Specifically, the response ranking optimizer re-ranks the results and optimizes the ranking of multiple recommendations / alternative devices / descriptions returned by the large model. If the model outputs multiple candidate results (such as recommending three device models), the following priority is prioritized:

[0143] High: matches user preferences (brand / package / voltage);

[0144] Medium: The option with a higher historical confirmation frequency;

[0145] Low: Recommends the option with a high logical score (assisted by the logit probability returned by the model). The logit is the "confidence score" output by the model and has not yet been converted into a final probability. The higher the logit, the more "biased" the model is towards this result.

[0146] The sorting results are displayed in the assistant window and can be marked as "matching preferences";

[0147] Specifically, the memory database manager is used to build long-term user preference modeling and behavioral memory storage, allowing the local AI assistant to have intelligent behavior with memory, providing the ability to locally persist memory content, and providing viewing, clearing, and exporting interfaces. All memory data is saved uniformly through a SQLite database file, and different data tables are established to record the user's static identity information or set preferences. It supports clearing memory and exporting, such as restoring default settings and exporting preference summaries. Administrators can also lock certain preference policies (such as naming encapsulation, etc.).

[0148] It also includes a multi-window parallel view module, in which a main table view, a dialogue window, an approval window and a graphic window are set;

[0149] The data in the main table view, dialogue window, approval window and graphic window are linked to each other. Clicking a field will automatically link and update other windows.

[0150] The present invention also provides a collaborative BOM data analysis and verification method based on an AI model, comprising the following steps:

[0151] S1: When a hardware BOM file is opened, the AI assistant window is automatically activated and the parsed file is read to initialize the user session. The user's memory configuration field preferences, operation habits, and project context are read to generate a unified data structure for the interactive module to call.

[0152] S2: Marking of documents through three triggering methods: user questions, field clicks, and automatic checks;

[0153] S3: Generate model inputs through the Prompt Builder and dynamically construct customized prompts based on the combined field content, user preferences, and project memory;

[0154] S4: local large language model operation and adjusted model parameters operation;

[0155] S5: The AI assistant generates answers including field explanations, verification judgments, modification suggestions, and recommended alternative devices. Responses are streamed out in the form of Streaming Tokens, allowing users to interrupt generation and re-ask.

[0156] S6: The user approval mechanism is triggered, and the "confirm / reject" operation is executed through buttons or dialogues. The user operation triggers the behavior collection module to record the approval result. The approval result is marked in the hardware BOM file and written into the audit log. The user's operation behavior and preference trends are recorded to extract long-term structural preferences.

[0157] S7: Subsequent conversations refer to the memory for semantic optimization and response sorting. The response order is optimized based on historical behavior, user profiles are constructed, and long-term collaborative efficiency optimization is achieved.

[0158] S8: Application export approval log, personalized preference log, result file model record file.

[0159] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0160] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. The collaborative BOM data analysis and verification system based on AI model is characterized by: include: An intelligent document perception and structure analysis module, which monitors and identifies the manifest file and converts the output into a document in a unified format; An AI assistant dialogue engine module, which uses a language model to interact with engineers to analyze documents in a unified format field by field, thereby assisting engineers in understanding the contents of the manifest file; A semantic classification and field specification module, which marks semantically ambiguous fields in a unified format document and triggers the AI assistant dialogue engine module to issue a suggestion message for the engineer to confirm; An intelligent verification and abnormality diagnosis module, which interfaces with fields in a device database, verifies and diagnoses a unified format document based on the fields in the device database, and generates corresponding suggestion records for engineers to process; A deployment and resource scheduling optimization module embeds and integrates the AI assistant dialogue engine module and the language model, and the embedded integrated AI assistant dialogue engine module analyzes unified format documents in an offline and low-power environment.

2. The collaborative BOM data analysis and verification system based on the AI model according to claim 1 is characterized by: The AI assistant dialogue engine module includes a user interaction layer, a semantic intermediate processing layer, and a reasoning and response layer; The user interaction layer uses PySide2 to provide users with a natural language interaction window; The semantic intermediate processing layer parses the unified format document through the session state manager, prompt builder and instruction intent recognizer; The reasoning and response layer uses the model service interface and the streaming output manager to dynamically add tags to the parsed unified format documents in real time.

3. The collaborative BOM data analysis and verification system based on the AI model according to claim 2 is characterized by: The model service interface uses the mistral.cpp framework to load the language model into the local machine in GGUF format, and uses INT4 quantization to optimize inference efficiency. The model service interface automatically switches and hot loads language models based on the context or task type of the unified format document.

4. The collaborative BOM data analysis and verification system based on the AI model according to claim 2 is characterized by: The streaming output manager pushes the step-by-step token results returned by the language model to the AI assistant dialogue engine module through SSE based on the Streaming Token inference mechanism; The AI assistant dialogue engine module dynamically adds tags in real time based on the level-by-level token results.

5. The collaborative BOM data analysis and verification system based on the AI model according to claim 1 is characterized by: The deployment and resource scheduling optimization module includes a model operation interface layer and an inference scheduling and caching layer; The model operation interface layer includes using a model loader to load the GGUF format weight file of the local language model into the memory, using a memory mapping mechanism to achieve efficient access and on-demand retrieval of the weight file, and building the local inference engine compiled from the language model into a static link library and embedding it into the desktop application front-end program; The inference scheduling and caching layer includes a thread pool scheduling mechanism, which encapsulates each inference request as an asynchronous task object and submits it to the thread pool managed by ThreadPoolExecutor for concurrent execution. After all requests are queued, they are scheduled and processed according to task priority.

6. The collaborative BOM data analysis and verification system based on the AI model according to claim 5 is characterized by: The deployment and resource scheduling optimization module also includes a model management and upgrade subsystem layer and a local log and authority control subsystem layer; The model management and upgrade subsystem layer includes automatically searching for language model files that meet the format through the specified model folder when the user switches multiple language models. When dynamically replacing the currently loaded language model, it unloads the current model weights and clears the related memory mapping through the resource release interface, then reloads the new GGUF model file, and refreshes the inference context cache and user session records. The local log and authority control subsystem layer is used to write each round of conversation records into the local database.

7. The collaborative BOM data analysis and verification system based on the AI model according to claim 6 is characterized by: It also includes an approval mechanism and a status tracking module, which establishes an auxiliary decision-making mechanism based on combining the suggestion information issued by the AI assistant dialogue engine module with the user's engineering behavior; The approval mechanism and status tracking module includes a suggested status data structure submodule, an approval interaction interface submodule, and a status mark visualization submodule; The suggestion status data structure submodule records and controls the entire life cycle of AI-generated suggestions, records all changes in the local log database, and forms a history of suggestion status changes. The approval interaction interface submodule provides a button operation or dialogue instruction triggering mechanism, allowing users to trigger state change behavior by clicking a button or using natural language instructions; The status mark visualization submodule is used to intuitively present the current processing status of each suggestion in the human-computer interaction interface.

8. The collaborative BOM data analysis and verification system based on the AI model according to claim 7 is characterized by: It also includes a memory mechanism and a personalized adaptation module, which are used to train the AI assistant in the AI assistant dialogue engine module; The memory mechanism and personalized adaptation module are used to track and record the user's feedback behavior on the system-generated suggestions, form a data basis for preference learning and behavior modeling, automatically inject user preferences and historical context into the large model prompt words, sort the results, and optimize the sorting of multiple suggestions or alternative devices or instructions returned by the large model, and build a long-term preference modeling and behavior memory storage for users, so that the AI assistant in the training AI assistant dialogue engine module has intelligent behavior of memory.

9. The collaborative BOM data analysis and verification system based on the AI model according to claim 8 is characterized by: It also includes a multi-window parallel view module, wherein the multi-window parallel view module is provided with a main table view, a dialogue window, an approval window and a graphic window; The data in the main table view, dialogue window, approval window and graphic window are linked to each other, and clicking on a field automatically links other windows to update.

10. A collaborative BOM data analysis and verification method based on an AI model is characterized by: The following steps are involved: S1: When a hardware BOM file is opened, the AI assistant window is automatically activated and the parsed file is read to initialize the user session. The user's memory configuration field preferences, operation habits, and project context are read to generate a unified data structure for the interactive module to call. S2: Marking of documents through three triggering methods: user questions, field clicks, and automatic checks; S3: Generate model inputs through the Prompt Builder and dynamically construct customized prompts based on the combined field content, user preferences, and project memory; S4: local large language model operation and adjusted model parameters operation; S5: The AI assistant generates answers including field explanations, verification judgments, modification suggestions, and recommended alternative components. Responses are streamed out in the form of Streaming Tokens, allowing users to "interrupt generation" and "ask again." S6: The user approval mechanism is triggered, and the "confirm / reject" operation is executed through buttons or dialogues. The user operation triggers the behavior collection module to record the approval result. The approval result is marked in the hardware BOM file and written into the audit log. The user's operation behavior and preference trends are recorded to extract long-term structural preferences. S7: Subsequent conversations refer to the memory for semantic optimization and response sorting. The response order is optimized based on historical behavior, user profiles are constructed, and long-term collaborative efficiency optimization is achieved. S8: Application export approval log, personalized preference log, result file model record file.

Images