Bidding document review method and system

Abstract

The invention provides a bidding document review method and system. The method comprises the steps of obtaining a to-be-reviewed bidding document text and a corresponding bidding document text; at least one type of business compliance auditing prompt words, technical compliance auditing prompt words, qualification matching auditing prompt words and format normalization auditing prompt words is obtained, and each type of auditing prompt words is designed based on bid invitation document terms in advance; the bidding information at least comprises an auditing type, a bidding clause / format, bidding content, problem description, a risk level, a clause basis, a modification suggestion and influence evaluation; based on the obtained auditing prompt word and the structured text of the bidding document, analyzing the structured text of the bidding document to be audited, and integrating multi-category analysis results to generate an analysis report; according to the bidding document review method and system provided by the invention, the review accuracy is improved through multi-dimensional deep review, and more powerful support is provided for intelligent development of the bidding field.

Description

Technical Field

[0001] This invention relates to the field of electronic bidding, and more specifically, to a method and system for reviewing bid documents. Background Technology

[0002] With the development of computer and internet technology, bidding has become a mainstream procurement method in today's market economy. The preparation and review of bid documents during the bidding process are highly specialized, typically relying on manual review of commercial terms, technical responses, qualifications, and format compliance, among other aspects. This process suffers from problems such as low review efficiency, low accuracy, and strong subjectivity.

[0003] Chinese invention patent CN106503930A discloses a document review method and apparatus. This method retrieves general concerns corresponding to the tender document from a server-side indicator database. Based on these general concerns, the tender document is scanned word by word to obtain the scanned result. Then, based on the scanned result and the entire content of the tender document, specific concerns for the user and corresponding remarks for each specific concern are derived. Since these specific concerns and their corresponding remarks indicate the review requirements of the tender document, the bid document can be reviewed based on these specific concerns and their corresponding remarks, resulting in suggested modifications. This allows bidders to make targeted modifications to their bid documents to ensure they meet the review requirements of the tender document.

[0004] However, while the above methods reduce the workload of bidders and tendering parties to some extent and improve work efficiency, they cannot conduct in-depth reviews of tender documents, accurately and efficiently screen out bidders that meet the tendering requirements, and it is also difficult to quantify the review results. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a method and system for reviewing tender documents based on intelligent agents. By combining specialized intelligent agents to conduct in-depth reviews of tender documents, the efficiency and accuracy of tender document review are improved, providing tenderers with more scientific and reliable decision support.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a method for reviewing tender documents, comprising: Obtain the text of the tender documents to be reviewed and the corresponding text of the tender documents; Obtain at least one of the following categories of audit prompts: business compliance audit prompts, technical compliance audit prompts, qualification matching audit prompts, and format standardization audit prompts. Each category of audit prompts should be designed in advance based on the terms of the tender documents and should include at least the audit type, tender terms / format, tender content, problem description, risk level, basis of the terms, modification suggestions, and impact assessment. Based on the obtained audit prompts and the structured text of the tender documents, the structured text of the tender documents to be reviewed is analyzed, and the analysis results from multiple categories are integrated to generate an analysis report.

[0007] As can be seen, this solution pre-designs multiple types of audit prompts based on the terms of the tender documents. Subsequently, based on each type of audit prompt and the structured text of the tender documents, the structured text of the tender documents to be reviewed is analyzed to ensure that the documents are rigorously examined from commercial substance, technical feasibility, subject qualification to external format. The analysis results from multiple perspectives are integrated to generate an analysis report, making the review results more reliable and providing the tendering party with more accurate decision-making basis. Furthermore, the prompts themselves constitute a set of review standards, ensuring that the review quality of documents remains consistent across different personnel at different times.

[0008] In addition, unlike directly reviewing the original tender documents, this solution reviews the text of the tender documents to be reviewed. The structured text makes the logic of the document content clearer, and the intelligent agent can analyze it based on this, which can more accurately identify problems and reduce erroneous judgments caused by ambiguity in text understanding.

[0009] In a preferred technical solution, based on the obtained audit prompts and the structured text of the tender documents, the structured text of the tender documents to be reviewed is analyzed to obtain an analysis report, including: Based on the business compliance audit prompts, verify whether the business terms in the structured text of the tender documents comply with the business requirements of the tender documents; the business terms should at least include the tender validity period, payment terms, liability for breach of contract, delivery time, and after-sales service; Based on the technical compliance audit prompts, verify whether the technical solutions in the structured text of the tender documents comply with the technical specifications of the tender documents. The tender technical specifications include at least core performance indicators, materials and processes, quality standards, environmental adaptability, safety, and technical compliance. Based on the qualification matching verification prompts, check whether the qualification content in the structured text of the tender documents meets the qualification requirements of the tender documents. The qualification content includes at least the enterprise qualification, performance certificate, personnel certificates and security deposit. Based on the format compliance review prompts, check whether the structured text of the tender document meets the format requirements of the tender document. The format must at least include a table of contents, signatures, page numbers, numbering, and completeness of attachments. Prompts guide the agent to conduct targeted reviews, clarifying the review focus and direction, avoiding blind analysis, and thus improving review efficiency. Since prompts are generated based on the specific requirements of the tender documents, the agent's review based on these prompts better aligns with the tender requirements, improving the match between review results and actual requirements, making the review results more reliable, reducing review errors, and mitigating the bidding risks caused by inaccurate reviews.

[0010] In a preferred technical solution, the generation of the analysis report includes: after obtaining the analysis results of multiple categories, integrating and deduplicating the problem risk points in the analysis results of each category, and calculating the risk score according to the risk level of each category and the weight of each category of risk.

[0011] By integrating and deduplicating the results of multi-category analysis and dynamically assigning weights to the importance of different categories, the interference of duplicate information on the analysis results is avoided, and the risk scoring is made more in line with the actual situation of bidding. The generated analysis report provides a reliable basis for decision-makers.

[0012] In a preferred technical solution, integrated prompts are designed in advance based on the evaluation requirements, including at least integration and deduplication, risk scoring, sorting output, generation of overall evaluation, and output format. Based on the integrated prompts, the problem risk points in the multi-category analysis results are integrated and deduplicated. The problem risk points include at least the risk type, bidding terms, bidding content, problem description, risk level, basis of terms, modification suggestions, and impact assessment.

[0013] During the integration and deduplication process, more targeted prompts are designed to help the agent identify and remove duplicate information, resolve mutually exclusive judgments on the same clause, and ensure that no problem or risk point is overlooked, making the analysis report more comprehensive and accurate. In the risk scoring calculation process, targeted weighting standards can more realistically reflect the overall risk status of the tender documents, providing support for generating a high-quality overall evaluation and helping to improve the efficiency and overall quality of the analysis report generation.

[0014] In a preferred technical solution, the method for obtaining the text of the tender document to be reviewed and the corresponding tender document text is as follows: The original documents of the tender documents and bidding documents are obtained separately. The original documents are then formatted and checked. Based on the check results, the corresponding file parsing tools are called to parse the documents. The original document types include at least one of the following formats: PDF, DOCX, DOC, TXT, and rich text. Based on the logical hierarchy of documents, chapters, clauses, tables, and metadata, the parsed results are converted into structured text; After obtaining the detection results, the process checks whether the original document contains scanned copies or images. If it does, an OCR tool is used to recognize the text in the scanned copy or image and convert it into searchable text. Then, based on the detection results, the corresponding file parsing tool is called to parse the document.

[0015] By using file parsing and OCR tools, the system enables rapid detection and conversion of document types, eliminating the need for manual judgment and processing and improving the efficiency of file processing. It can adapt to various original document types, including scanned documents and images, expanding the range of files that can be processed. This allows the system to effectively process tender documents and bidding documents of different formats, enhancing the system's applicability and flexibility.

[0016] In a preferred technical solution, after obtaining the analysis results, a document highlighting tool is used to map the analysis results to the corresponding positions in the text of the tender document to be reviewed, according to paragraph index, page number, line number, or OCR coordinates, and the corresponding positions are highlighted.

[0017] Highlighting makes the review results immediately visible in the original document, facilitating review by relevant personnel and eliminating the need to search for issues within large amounts of text, thus greatly improving the visualization of the review results. Quickly locating issues saves review time and improves review efficiency, especially when dealing with lengthy tender documents, helping reviewers quickly focus on problems and accelerate the review process.

[0018] Secondly, the present invention provides a tender document review system, comprising: It includes a user layer, a professional agent layer, and an output layer; The user layer is used to input the text of the tender document to be reviewed and the corresponding tender document text, and to obtain at least one of the following: business compliance review prompts, technical compliance review prompts, qualification matching review prompts, and format standardization review prompts. Each type of review prompt is pre-designed based on the tender document clauses and includes at least the review type, tender clauses / formats, tender content, problem description, risk level, clause basis, modification suggestions, and impact assessment. The professional intelligent agent layer is equipped with an intelligent agent module, which is used to analyze the structured text of the tender documents to be reviewed based on the obtained review prompt words and the structured text of the tender documents, and integrate the analysis results of multiple categories to generate an analysis report. The output layer is used to output analysis reports.

[0019] In a preferred embodiment, the intelligent agent module includes: A business compliance intelligent agent is used to review business compliance prompts and verify whether the business terms in the structured text of the tender documents comply with the business requirements of the tender documents; the business terms include at least the bid validity period, payment terms, liability for breach of contract, delivery time, and after-sales service; A technology compliance intelligent agent is used to verify whether the technical solutions in the structured text of the tender documents comply with the technical specifications of the tender documents based on technology compliance audit prompts. The tender technical specifications include at least core performance indicators, materials and processes, quality standards, environmental adaptability, safety, and technical compliance. The qualification matching intelligent agent is used to check whether the qualification content in the structured text of the tender document meets the qualification requirements of the tender document based on qualification matching review prompts. The qualification content includes at least enterprise qualifications, performance certificates, personnel certificates and security deposit. A format specification intelligent agent is used to check whether the format of the structured text of the tender document meets the requirements of the tender document format based on the format specification review prompt words. The format includes at least a table of contents, signatures, page numbers, numbering, and completeness of attachments. The integrated agent is used to integrate the problem descriptions and modification suggestions output by the above agents to generate an analysis report, and to calculate a risk score based on the risk level and the weight of each risk category.

[0020] It can be seen that different intelligent agents focus on different aspects of the review. For example, the business compliance intelligent agent focuses on business terms, while the technology compliance intelligent agent focuses on technical solutions. They cooperate with each other to conduct a comprehensive review of the tender documents from multiple dimensions.

[0021] The results integration agent integrates the analysis results of each agent, calculates the compliance score based on the importance of different chapters and clauses of the tender documents, and generates a detailed analysis report, providing the tendering party with intuitive and accurate decision-making basis.

[0022] In a preferred embodiment, the technology further includes an MCP service layer, an A2A protocol layer, and a coordinator layer, wherein the MCP service layer includes: The text conversion module is used to perform format detection on the original documents of the tender documents and bidding documents, and call the corresponding file parsing tool to parse the documents based on the detection results; the original document type includes at least one of PDF, DOCX, DOC, TXT, and rich text format; the parsing results are converted into structured text according to the logical hierarchy of documents, chapters, clauses, tables and metadata; OCR tools are used to recognize and convert text in scanned documents or images into searchable text. A document highlighting tool used to insert highlighted annotations at specified locations and output them in PDF / DOCX format; The fault-tolerance module is used to correct non-standard analysis results. The correction includes at least pre-cleaning, layer-by-layer regularization, and structured or manual review. The non-standard results include missing quotation marks, mismatched brackets, and mixed use of single quotation marks. The A2A protocol layer is used to define the message format and state machine for exchanging tasks between intelligent agents. The message format and state machine include JSON format for task requests / responses, status reports, error codes and retry mechanisms, as well as authentication and access control. The coordinator layer includes a coordinator that receives the original documents of tender documents and bidding documents. If the original documents contain scanned copies or images, the coordinator uses an OCR tool to recognize and convert the text in the scanned copies or images into searchable text. It performs format checks on the original documents of tender documents and bidding documents, and calls the corresponding file parsing tool to parse the documents based on the check results. It also uses an A2A protocol to schedule business compliance, technical compliance, and qualification matching agents to analyze the structured text of the tender documents to be reviewed, and calls an integration agent to integrate and optimize problem descriptions and modification suggestions. It calls a fault-tolerance module to correct non-standard analysis results and generate standardized analysis results. Finally, it calls a file highlighting tool to map the analysis results to the corresponding positions in the text of the tender documents to be reviewed according to paragraph index, page number, line number, or OCR coordinates, and highlights the corresponding positions to generate a tender document with highlighted annotations.

[0023] This invention has outstanding substantive features and significant progress compared to the prior art, and has the following beneficial effects: (1) Design multiple types of audit prompts in advance based on the terms of the tender documents. Then, based on each type of audit prompt and the structured text of the tender documents, analyze the structured text of the tender documents to be reviewed to ensure that the documents are rigorously examined from commercial substance, technical feasibility, subject qualification to external format. Integrate the analysis results from multiple perspectives to generate an analysis report, making the review results more reliable and providing the tendering party with more accurate decision-making basis.

[0024] (2) Based on the logical hierarchy of chapters and clauses, the document is converted into structured text, multiple intelligent agents are used for parallel review, and targeted analysis is performed based on prompt words, which significantly improves the efficiency and accuracy of review. This can save a lot of time and energy for bidding and tendering work, while overcoming the subjective risk of manual review.

[0025] (3) Based on the importance of different chapters and clauses of the tender documents, the compliance score is calculated by weighting, and the analysis results are written back to the tender documents in the form of annotations and highlights, providing an intuitive and comprehensive basis for the bidder's manual secondary correction and the tenderer's decision-making.

[0026] (4) The modules of the tender document review system work together to adapt to the bidding needs of more industries and scenarios. The coordinator module and the fault tolerance module further ensure the efficient and stable operation of the system, providing stronger support for the intelligent development of the bidding field. Attached Figure Description

[0027] Figure 1 This is a flowchart of the tender document review method of the present invention; Figure 2 This is a flowchart of the tender document review system of the present invention; Figure 3 This is an architecture diagram of the tender document review system of the present invention. Detailed Implementation

[0028] The technical solution of the present invention will be further described in detail below through specific embodiments.

[0029] To facilitate understanding of the technical solutions provided in this application, the technical terms involved in the embodiments of this application are explained below.

[0030] As an advanced artificial intelligence entity, an intelligent agent achieves its preset goals by continuously perceiving the external environment, making autonomous decisions, and executing actions.

[0031] MCP (Model Context Protocol) is an open-source protocol launched by Anthropic in 2024. It aims to enable seamless integration of large language models (LLMs) with external data sources and tools, and to establish a secure, bidirectional link between large models and data sources.

[0032] The A2A Protocol (Agent-to-Agent Protocol) is an open protocol jointly launched by Google and more than 50 technology service providers, including SAP, ServiceNow, and LangChain, in 2025. It aims to provide a common communication language and collaboration mechanism for intelligent agents, enabling seamless interaction and collaboration between different intelligent agents.

[0033] Example 1 In this embodiment of the application, a method for reviewing tender documents is proposed, including: Obtain the text of the tender documents to be reviewed and the corresponding tender documents.

[0034] Specifically, the method for obtaining the text of the tender documents to be reviewed and the corresponding text of the tender documents is as follows: The original documents of the tender documents and bidding documents are obtained separately. The original documents are then formatted and checked. Based on the check results, the corresponding file parsing tools are called to parse the documents. The original document types include at least one of the following formats: PDF, DOCX, DOC, TXT, and rich text. Specifically, the file parsing process includes: file type detection, OCR recognition of scanned pages, extraction of paragraphs / page numbers / tables / headings, encoding repair, and preprocessing, such as removing special characters and blank lines; the parsing results can be seen in the following example: Page-level segmentation (page id, page no) Paragraph-level segmentation (para id, start offset, end offset) Clause / Title Recognition (section id, depth, title text) Extract the table as a structured table object (table id, rows, cols); Based on the logical hierarchy of documents, chapters, clauses, tables, and metadata, the parsed results are converted into structured text; preferably, the structured text is in JSON format, as shown in the following sample: "doc id": "file-001" "sections":[{"section id":"sec-1","title":"Chapter 1 Project Overview","page":1, "clauses":[{"clause_id":"cl","text":"Article 1: Delivery Date.."}]}, {"section id":"sec-2","title":"Chapter 2 Technical Requirements","page":3, "clauses":[{"clause id":"c2","text":"Item 2: Performance Parameters…"}]} ], "tables":[{"table_id":"tl","page":2,"rows" :5}], "meta": {"encoding" :"utf-8","pages":10} } After obtaining the detection results, the process checks whether the original document contains scanned copies or images. If it does, an OCR tool is used to recognize the text in the scanned copy or image and convert it into searchable text. Then, based on the detection results, the corresponding file parsing tool is called to parse the document.

[0035] Obtain at least one of the following categories of audit prompts: business compliance audit prompts, technical compliance audit prompts, qualification matching audit prompts, and format standardization audit prompts. Each category of audit prompts should be designed in advance based on the terms of the tender documents and should include at least the audit type, tender terms / format, tender content, problem description, risk level, basis of the terms, modification suggestions, and impact assessment.

[0036] In one embodiment, the business compliance agent prompts are constructed based on the business terms of the tender documents, enabling the business compliance agent to meticulously review the commercial compliance of the tender documents.

[0037] Examples of business compliance intelligent agent prompts are as follows: "You are a professional business compliance review expert, responsible for comprehensively reviewing the commercial compliance of tender documents, starting from the commercial terms of the tender documents. You need to:" 1. Carefully review the commercial terms in the tender documents, examining each clause of the tender documents. 2. Identify any business details that are inconsistent with or missing from the tender requirements. 3. Assess the severity of each problem (high / medium / low) 4. Citify relevant clauses from the tender documents as the basis. 5. Provide specific rectification suggestions to ensure that the tender documents fully comply with the tender's commercial requirements. The output format must be a JSON array, and each question must contain the following fields: {"Type":"Business Compliance Risk", "Tender Terms": "Original text of relevant terms in the tender documents", "Tender Contents": "Corresponding content in the tender documents", Problem Description: "Specific Inconsistencies", Risk Level: "High / Medium / Low" Basis: "Tender Document Clause Number / Description", Suggested modifications: "How to modify or supplement this document?" Impact Assessment: "The impact of this issue on the evaluation results or the qualification to win the bid." Special note: 1. Pay close attention to the commercial terms, including bid validity period, payment terms, liability for breach of contract, delivery time, and after-sales service. 2. Identify any significant deviations that may lead to rejection of the bid. 3. Check for any omissions, ambiguities, or clauses that conflict with the tender requirements. 4. Ensure the recommendations are actionable and compliant. The technical compliance agent prompts are constructed based on the technical specifications in the tender documents, enabling the technical compliance agent to meticulously review the compliance of the technical solutions in the tender documents.

[0038] Examples of technology-compliant agent prompts are as follows: "You are a professional technical compliance review expert, responsible for comprehensively reviewing the technical solutions in the tender documents based on the technical specifications in the tender documents. You need to:" 1. Compare with the technical requirements and standards in the tender documents. 2. Identify the parts of the tender documents that do not meet the technical specifications, performance requirements, and standards. 3. Assess the severity (high / medium / low) of each technical deviation. 4. Use the technical specifications in the tender documents as a basis. 5. Provide improvement suggestions to ensure that the technical solution fully meets the bidding requirements. The output format must be a JSON array, and each question must contain the following fields: {"Type":"Technology Compliance Risk", "Technical Terms and Conditions in the Tender Documents": "Original Text of Relevant Technical Terms and Conditions in the Tender Documents" "Technical Content of Bidding": "Corresponding technical description in the bidding documents", Problem Description: "Specific technical details of the non-compliance", Risk Level: "High / Medium / Low" "Based on": "Technical Specification Number / Standard of the Tender Document", "Suggested modifications": "How should the technical solution be adjusted?" Impact Assessment: "Impact on performance, acceptance, or evaluation" Special note: 1. Focus on core performance indicators, materials and processes, quality standards, environmental adaptability, and safety. 2. Identify technical designs that conflict with national or industry standards. 3. Check for any uncovered or omitted technical requirements. 4. Ensure that the recommendations are achievable within the scope of the tender. The qualification matching intelligent agent prompts are set according to the qualification requirements of the bidding documents, enabling the qualification matching intelligent agent to efficiently verify the qualification matching of the bidding documents.

[0039] Examples of prompts for competency-matching agents are as follows: "You are a professional qualification review expert, responsible for comprehensively verifying the qualification matching of bid documents based on the qualification requirements in the tender documents. You need to:" 1. Compare the qualification requirements for bidders with those specified in the tender documents. 2. Verify the completeness, validity, and compliance of the qualification certificates in the tender documents. 3. Identify any missing, expired, or non-compliant qualification documents. 4. Assess the risk level (high / medium / low) of each issue. 5. Provide suggestions for supplementing or replacing. The output format must be a JSON array, and each question must contain the following fields: {"Type":"Qualification Matching Risk", "Bidding Qualification Requirements": "Original text of relevant qualification clauses in the bidding documents", "Bidding Qualification Materials": "The qualification materials corresponding to the bidding documents", Problem Description: "Specific areas where qualifications are not met", Risk Level: "High / Medium / Low" Basis: "Qualification Requirements Clause in the Tender Document" Suggested modifications: "How to supplement or replace" Impact Assessment: "The impact of this issue on qualification review and bid validity" Special note: 1. Verify the validity period, issuing authority, level, and professional scope of the qualification certificate. 2. Check whether the project-specific registration, licensing, and certification requirements are met. 3. Identify any defects that could lead to disqualification. 4. Ensure that the recommendations can be implemented before the bid deadline. The format-standard intelligent agent prompts are set according to the preparation and format requirements of the bidding documents, enabling the format-standard intelligent agent to efficiently check the format compliance of the bidding documents.

[0040] The following are examples of formatted intelligent agent prompts: "You are a professional expert in reviewing bid document formats and specifications, responsible for comprehensively examining bid documents based on their preparation and format requirements. You need to:" 1. Refer to the preparation instructions, format template, and page numbering requirements of the tender documents. 2. Check whether the layout, chapter order, table of contents, signatures, attachments, etc. meet the requirements. 3. Identify any significant formatting deviations that could lead to rejection of the bid. 4. Assess the risk level (high / medium / low) of each issue. 5. Provide revision suggestions to ensure the document format fully complies with the bidding requirements. The output format must be a JSON array, and each question must contain the following fields: {"Type":"Formatting Standards Risk", "Tender Format Requirements": "Original Text of Relevant Format Clauses in Tender Documents", "Tender Document Contents": "The corresponding format content in the tender documents", Problem Description: "Specific formatting errors", Risk Level: "High / Medium / Low" "Based on": "Tender Document Format Requirements Clause", Suggested edits: "How to adjust the formatting?" Impact Assessment: "The impact of this issue on the validity or scoring of the bid." Special note: 1. Check whether the template and font size required by the tender documents have been used. 2. Confirm that the table of contents, chapter numbers, and attachment numbers are consistent with the bidding requirements. 3. Check the completeness and position of the signatures and seals. 4. Ensure that the recommendations can be implemented quickly and do not change the substantive content of the document.

[0041] Finally, based on the obtained audit prompts and the structured text of the tender documents, the structured text of the tender documents to be reviewed is analyzed, and the analysis results from multiple categories are integrated to generate an analysis report.

[0042] As can be seen, the above scheme pre-designs multiple types of audit prompts based on the terms of the tender documents. Subsequently, based on each type of audit prompt and the structured text of the tender documents, the structured text of the tender documents to be reviewed is analyzed to ensure that the documents are rigorously examined from commercial substance, technical feasibility, subject qualification to external format. The analysis results from multiple perspectives are integrated to generate an analysis report, making the review results more reliable and providing the tendering party with a more accurate basis for decision-making.

[0043] Example 2 like Figure 1 As shown in the embodiments of this application, a specific implementation method for generating an analysis report is provided.

[0044] Based on business compliance audit prompts, technical compliance audit prompts, qualification matching audit prompts, format standardization audit prompts, and the structured text of the tender documents, the structured text of the tender documents to be reviewed is analyzed to obtain an analysis report, including: Based on the business compliance audit prompts, verify whether the business terms in the structured text of the tender documents comply with the business requirements of the tender documents; the business terms should at least include the tender validity period, payment terms, liability for breach of contract, delivery time, and after-sales service; Based on the technical compliance audit prompts, verify whether the technical solutions in the structured text of the tender documents comply with the technical specifications of the tender documents. The tender technical specifications include at least core performance indicators, materials and processes, quality standards, environmental adaptability, safety, and technical compliance. Based on the qualification matching verification prompts, check whether the qualification content in the structured text of the tender documents meets the qualification requirements of the tender documents. The qualification content includes at least the enterprise qualification, performance certificate, personnel certificates and security deposit. Based on the format compliance review prompts, check whether the format of the structured text of the tender document meets the requirements of the tender document format. The format should at least include a table of contents, signatures, page numbers, numbering, and completeness of attachments.

[0045] In one embodiment, after obtaining the analysis results, a document highlighting tool is used to map the analysis results to the corresponding positions in the text of the tender document to be reviewed, according to paragraph index, page number, line number, or OCR coordinates, and the corresponding positions are highlighted.

[0046] Specifically, the reviewed tender documents are annotated and highlighted, and the text formats include PDF, DOC, and DOCX, with different colored markers to distinguish the risk level.

[0047] In one embodiment, after obtaining the multi-category analysis results, the problem risk points in each category analysis results are integrated and deduplicated, and a risk score is calculated based on the risk level of each category and the weight of each category of risk.

[0048] Specifically, the process of integrating and deduplicating risk points from the analysis results of each category includes manual screening and simple algorithm deduplication. The risk score is calculated by assigning corresponding weights to the importance of each risk category using a weighted average method, then multiplying the risk level by the weight, and finally summing all the products to obtain the risk score.

[0049] Specifically, after obtaining the multi-category analysis results, before integrating and deduplicating the problem risk points in each category of analysis results or calculating risk scores, non-standard analysis results are also corrected. The correction includes at least pre-cleaning, layer-by-layer regularization and structuring or manual review; the non-standard results include missing quotation marks, mismatched parentheses, and mixed use of single quotation marks.

[0050] In one embodiment, integrated prompts are designed in advance based on the evaluation requirements, including at least integration and deduplication, risk scoring, sorting output, generating overall evaluation, and output format. Based on the integrated prompts, the problem risk points in the multi-category analysis results are integrated and deduplicated. The problem risk points include at least the risk type, bidding terms, bidding content, problem description, risk level, basis of terms, modification suggestions, and impact assessment.

[0051] In one embodiment, an example of integrating prompt words is as follows: "You are an integration analyst responsible for the final review of tender documents. You will receive analysis results from four different professional agents:" 1. Business Compliance Agent 2. Technology Compliance Agent 3. Qualification matching agent 4. Format and Standardization of Agent Each Agent outputs a JSON array containing multiple risk points, with each risk point having the following fields: {"Type":"Risk Category", "Terms": "The original text of the specific terms", Problem Description: Risk Description Risk Level: "High / Medium / Low" "Legal basis or standard": "Relevant legal provisions or technical standards", "Suggested Revisions": "Specific suggested revisions", "Impact on our side": "Potential consequences of this issue" Your task: 1. Integration and Deduplication: Merge all risk points of the four agents and remove duplicate or highly similar entries.

[0052] 2. Risk Scoring: Calculate a risk score for each risk point. High risk = 3 points Medium risk = 2 points Low risk = 1 point At the same time, weights are assigned based on the importance of different agents: Business compliance agent weight: 0.35; Technology compliance agent weight: 0.30; Agent weight for matching qualifications: 0.20; Format specification Agent weight: 0.15; Individual risk deduction = Risk level score × Corresponding Agent weight 3. Sorting Output: Arrange risk points from highest to lowest according to risk deduction score.

[0053] 4. Generate overall evaluation: Summarize the scores of all risk points to obtain the total score (out of 100 – total deductions for all risks).

[0054] An overall evaluation will be given based on the total score: ≥85 points: Excellent 70-84 points: Good 50~69 points: medium <50 points: Significant risk exists. 5. Output format: The output must be a JSON string containing the following fields: {"Total Score":"Overall Score", Overall rating: "Excellent / Good / Average / Significant risk present" "Risk List":[ {"Type":"Risk Category", "Terms": "The original text of the specific terms", Problem Description: Risk Description Risk Level: "High / Medium / Low" "Legal basis or standard": "Relevant legal provisions or technical standards", "Suggested Revisions": "Specific suggested revisions", "Impact on our side": "Potential consequences of this issue", Special note: 1. You cannot overlook any risk point. 2. If different agents have differing opinions on the same clause, their descriptions should be combined and the highest risk level should be used. 3. The total score must be calculated to two decimal places.

[0055] By integrating the analysis results of prompts for business compliance review, technical compliance review, qualification matching review, and format standardization review, the intelligent agent can efficiently integrate and resolve conflicts in the analysis results of the tender documents, calculate risk scores, and output an analysis report. Preferably, the analysis report is a structured report in JSON format. This structured analysis report includes a list of issues, evidence fragments, and scores.

[0056] In some embodiments, modification suggestions can be annotated in the corresponding positions of the tender documents based on the integrated analysis results. The color of the annotation can be set according to the risk level, for example, red for high risk level, orange for medium risk level, and yellow for low risk level, so as to facilitate manual review and adjustment based on the analysis results later.

[0057] Example 3 like Figure 3As shown in the embodiments of this application, a tender document review system is proposed, including a user layer, a professional agent layer and an output layer; The user layer is used to input the text of the tender document to be reviewed and the corresponding tender document text, and to obtain at least one type of review prompts: business compliance review prompts, technical compliance review prompts, qualification matching review prompts, and format standardization review prompts. Each type of review prompt is pre-designed based on the tender document clauses and includes at least the review type, tender clauses / formats, tender content, problem description, risk level, clause basis, modification suggestions, and impact assessment. In some embodiments, the tender document can be the tender text, attachments, or forms. The tender document includes the tender announcement, technical specifications, and scoring criteria. In addition, the review focus can be selected, such as emphasizing technology / business / qualification. The professional intelligent agent layer is equipped with an intelligent agent module, which is used to analyze the structured text of the tender documents to be reviewed based on the obtained review prompt words and the structured text of the tender documents, and integrate the analysis results of multiple categories to generate an analysis report. The output layer is used to output analysis reports.

[0058] Specifically, users can upload tender documents (tender text, attachments, forms) and bidding documents (bidding announcement, technical specifications, scoring criteria) through the user-level front-end interface or API, and can choose the focus of review (such as emphasizing technology, business or qualifications).

[0059] In one embodiment, such as Figure 2 As shown, the intelligent agent module includes: A business compliance intelligent agent is used to review business compliance prompts and verify whether the business terms in the structured text of the tender documents comply with the business requirements of the tender documents; the business terms include at least the bid validity period, payment terms, liability for breach of contract, delivery time, and after-sales service; A technology compliance intelligent agent is used to verify whether the technical solutions in the structured text of the tender documents comply with the technical specifications of the tender documents based on technology compliance audit prompts. The tender technical specifications include at least core performance indicators, materials and processes, quality standards, environmental adaptability, safety, and technical compliance. The qualification matching intelligent agent is used to check whether the qualification content in the structured text of the tender document meets the qualification requirements of the tender document based on qualification matching review prompts. The qualification content includes at least enterprise qualifications, performance certificates, personnel certificates and security deposit. A format specification intelligent agent is used to check whether the format of the structured text of the tender document meets the requirements of the tender document format based on the format specification review prompt words. The format includes at least a table of contents, signatures, page numbers, numbering, and completeness of attachments. The integrated agent is used to integrate the problem descriptions and modification suggestions output by the above agents to generate an analysis report, and to calculate a risk score based on the risk level and the weight of each risk category.

[0060] Integrating intelligent agents can also resolve conflicts arising from mutually exclusive judgments of the same clause by different specialized intelligent agents.

[0061] The aforementioned solution, by employing multi-agent parallel collaboration and assigning tasks according to professional dimensions, can improve review efficiency and reduce professional blind spots. It also introduces a risk scoring mechanism, quantifying subjective judgments into comparable compliance scores based on weights, thus facilitating bidding decisions.

[0062] Specifically, intelligent agents complete analysis tasks through prompts; intelligent agents also include financial intelligent agents, risk and compliance intelligent agents, or legal advisory intelligent agents.

[0063] In one embodiment, various intelligent agents are constructed using a large language model or a local fine-tuning model, bidding documents, and various format and standardization review prompts, and the intelligent agents communicate through the A2A protocol.

[0064] In one embodiment, it also includes an MCP service layer, an A2A protocol layer, and a coordinator layer.

[0065] The MCP service layer includes: The text conversion module is used to perform format detection on the original documents of tender documents and bidding documents, and call the corresponding file parsing tool to parse the documents based on the detection results; the original document type includes at least one of PDF, DOCX, DOC, TXT, and rich text format; the parsing results are converted into structured text according to the logical hierarchy of documents, chapters, clauses, tables, and metadata; this setting can support robust document parsing with multiple formats and encodings, and improve the coverage of structured extraction; OCR tools are used to recognize and convert text in scanned documents or images into searchable text.

[0066] As can be seen, the MCP service, as a set of tools, provides reusable document and file operation interfaces to the knowledge provider and coordinator. The MCP service ensures a unified description and registration (tool table) for each tool, enabling various intelligent agents to invoke them using a consistent interface.

[0067] Specifically, the MCP service layer also includes decoding tools with multi-encoding capabilities, including UTF-8, GBK, GB2312, and other character encoding methods, supporting the encoding of simplified Chinese characters or multilingual characters. The MCP service layer provides base64 encoding to encode the output documents in the two mainstream formats, PDF and DOCX, for transmission via API.

[0068] This document highlighting tool is used to insert highlighted annotations at specified locations and output them in PDF / DOCX format. Specifically, the specified location refers to the corresponding position found in the source file based on paragraph indexes, page numbers, line numbers, or OCR coordinates from the analysis results. It's understandable that outputting the original document with highlighted annotations can improve the efficiency of manual review and shorten the delivery cycle.

[0069] The fault-tolerance module is used to correct non-standard analysis results. The correction includes at least pre-cleaning, layer-by-layer regularization and structured or manual review. The non-standard results include missing quotation marks, mismatched brackets, and mixed use of single quotation marks.

[0070] Specifically, the fault-tolerant module adopts a multi-mode parsing strategy, which includes pre-cleaning the pseudo-JSON output by the agent; attempting to recover valid JSON using layer-by-layer regular expressions and a structured parser; and returning the original text that failed to be parsed as a text field with instructions for manual review.

[0071] Specifically, in the data interaction and error handling between the integration agent and various specialized agents, the fault-tolerance module ensures that even if the output JSON format of a particular agent is not standardized (e.g., missing quotes, mismatched parentheses, or mixed use of single quotes), the integration agent can still successfully parse it and continue calculating the score, rather than the entire process crashing. This strategy is compatible with the output styles of different large models and reduces the task failure rate caused by parsing anomalies. The fault-tolerance mechanism makes the above solution applicable to different bidding scenarios and organizational sizes.

[0072] Specifically, the system supports horizontal scaling and load balancing of agents, and when some agents are unavailable, an uncertainty indicator is given in the report when an available result is returned.

[0073] In particular, operational and agent response logs are retained for each review in order to facilitate upgrades and model improvements.

[0074] The A2A protocol layer is used to define the message format and state machine for exchanging tasks between intelligent agents. The message format and state machine include JSON format for task requests / responses, status reports, error codes and retry mechanisms, as well as authentication and access control.

[0075] Specifically, the A2A protocol layer defines the JSON format for task requests / responses, including fields such as task_id, task_type, content, meta, and timeout; status reports include received, processing, completed, and failed; and error codes and retry mechanisms include a maximum number of retries and exponential backoff.

[0076] The coordinator layer includes a coordinator that receives the original documents of tender documents and bidding documents. If the original documents contain scanned copies or images, the coordinator uses an OCR tool to recognize and convert the text in the scanned copies or images into searchable text. It performs format checks on the original documents of tender documents and bidding documents, and calls the corresponding file parsing tool to parse the documents based on the check results. It also uses an A2A protocol to schedule business compliance, technical compliance, and qualification matching agents to analyze the structured text of the tender documents to be reviewed, and calls an integration agent to integrate and optimize problem descriptions and modification suggestions. It calls a fault-tolerance module to correct non-standard analysis results and generate standardized analysis results. Finally, it calls a file highlighting tool to map the analysis results to the corresponding positions in the text of the tender documents to be reviewed according to paragraph index, page number, line number, or OCR coordinates, and highlights the corresponding positions to generate a tender document with highlighted annotations.

[0077] In particular, the coordinator also supports asynchronous execution and result callbacks, and can perform task degradation when necessary, such as returning partial results and alerts when the Agent times out or becomes unreachable.

[0078] Example of an implementation scenario in this embodiment: Example 1: Self-inspection before bidding.

[0079] Users upload their tender documents and bidding technical specifications, and the system initiates automatic review. After multi-agent collaboration, a bid evaluation report is output, including the bid evaluation score, grade, and a risk list and solution suggestions regarding technical compliance, business compliance, qualification matching, and format standardization.

[0080] The system outputs an annotated PDF and a revision list, which users can use to make changes and re-upload for secondary review.

[0081] Example 2: Batch bid pool evaluation.

[0082] For large-scale projects, it is necessary to pre-screen multiple potential bids. The system imports bid documents in batches and automatically generates scores, providing a preliminary ranking based on the scores, which helps decision-makers prioritize resource allocation for key projects.

[0083] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation of the present invention or equivalent substitutions can be made to some technical features without departing from the spirit of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the scope of the technical solutions claimed in the present invention.

Claims

1. A method for reviewing tender documents, characterized in that, include: Obtain the text of the tender documents to be reviewed and the corresponding text of the tender documents; Obtain at least one of the following categories of audit prompts: business compliance audit prompts, technical compliance audit prompts, qualification matching audit prompts, and format standardization audit prompts. Each category of audit prompts should be designed in advance based on the terms of the tender documents and should include at least the audit type, tender terms / format, tender content, problem description, risk level, basis of the terms, modification suggestions, and impact assessment. Based on the obtained audit prompts and the structured text of the tender documents, the structured text of the tender documents to be reviewed is analyzed, and the analysis results from multiple categories are integrated to generate an analysis report.

2. The method for reviewing tender documents according to claim 1, characterized in that, The generation of the analysis report includes analyzing the structured text of the tender documents to be reviewed based on the obtained audit prompts and the structured text of the tender documents, and obtaining the analysis report, including: Based on the business compliance audit prompts, verify whether the business terms in the structured text of the tender documents comply with the business requirements of the tender documents; the business terms should at least include the tender validity period, payment terms, liability for breach of contract, delivery time, and after-sales service; Based on the technical compliance audit prompts, verify whether the technical solutions in the structured text of the tender documents comply with the technical specifications of the tender documents. The tender technical specifications include at least core performance indicators, materials and processes, quality standards, environmental adaptability, safety, and technical compliance. Based on the qualification matching verification prompts, check whether the qualification content in the structured text of the tender documents meets the qualification requirements of the tender documents. The qualification content includes at least the enterprise qualification, performance certificate, personnel certificates and security deposit. Based on the format compliance review prompts, check whether the format of the structured text of the tender document meets the requirements of the tender document format. The format should at least include a table of contents, signatures, page numbers, numbering, and completeness of attachments.

3. The method for reviewing tender documents according to claim 1, characterized in that, The generation of the analysis report includes: after obtaining the analysis results of multiple categories, integrating and deduplicating the problem and risk points in the analysis results of each category, and calculating the risk score according to the risk level of each category and the weight of each category of risk.

4. The method for reviewing tender documents according to claim 3, characterized in that, Design integrated prompts in advance based on the evaluation requirements, including at least integration and deduplication, risk scoring, sorting output, generation of overall evaluation, and output format; Based on the integrated prompts, the problem risk points in the multi-category analysis results are integrated and deduplicated. The problem risk points include at least the risk type, bidding terms, bidding content, problem description, risk level, basis of terms, modification suggestions, and impact assessment.

5. The method for reviewing tender documents according to claim 1, characterized in that, The method for obtaining the text of the tender documents to be reviewed and the corresponding text of the tender documents is as follows: The original documents of the tender documents and bidding documents are obtained separately. The original documents are then formatted and checked. Based on the check results, the corresponding file parsing tools are called to parse the documents. The original document types include at least one of the following formats: PDF, DOCX, DOC, TXT, and rich text. Based on the logical hierarchy of documents, chapters, clauses, tables, and metadata, the parsed results are converted into structured text; After obtaining the detection results, the process checks whether the original document contains scanned copies or images. If it does, an OCR tool is used to recognize the text in the scanned copy or image and convert it into searchable text. Then, based on the detection results, the corresponding file parsing tool is called to parse the document.

6. The bid document review method according to claim 3, characterized in that, include: After obtaining the analysis results, use the document highlighting tool to map the analysis results to the corresponding positions in the text of the tender document to be reviewed, according to paragraph index, page number, line number, or OCR coordinates, and highlight the corresponding positions.

7. A tender document review system, characterized in that, It includes a user layer, a professional agent layer, and an output layer; The user layer is used to input the text of the tender document to be reviewed and the corresponding tender document text, and to obtain at least one of the following: business compliance review prompts, technical compliance review prompts, qualification matching review prompts, and format standardization review prompts. Each type of review prompt is pre-designed based on the tender document clauses and includes at least the review type, tender clauses / formats, tender content, problem description, risk level, clause basis, modification suggestions, and impact assessment. The professional intelligent agent layer is equipped with an intelligent agent module, which is used to analyze the structured text of the tender documents to be reviewed based on the obtained review prompt words and the structured text of the tender documents, and integrate the analysis results of multiple categories to generate an analysis report. The output layer is used to output analysis reports.

8. A bid document review system according to claim 7, characterized in that, The intelligent agent module includes: A business compliance intelligent agent is used to review business compliance prompts and verify whether the business terms in the structured text of the tender documents comply with the business requirements of the tender documents; the business terms include at least the bid validity period, payment terms, liability for breach of contract, delivery time, and after-sales service; A technology compliance intelligent agent is used to verify whether the technical solutions in the structured text of the tender documents comply with the technical specifications of the tender documents based on technology compliance audit prompts. The tender technical specifications include at least core performance indicators, materials and processes, quality standards, environmental adaptability, safety, and technical compliance. The qualification matching intelligent agent is used to check whether the qualification content in the structured text of the tender document meets the qualification requirements of the tender document based on qualification matching review prompts. The qualification content includes at least enterprise qualifications, performance certificates, personnel certificates and security deposit. A format specification intelligent agent is used to check whether the format of the structured text of the tender document meets the requirements of the tender document format based on the format specification review prompt words. The format includes at least a table of contents, signatures, page numbers, numbering, and completeness of attachments. The integrated agent is used to integrate the problem descriptions and modification suggestions output by the above agents to generate an analysis report, and to calculate a risk score based on the risk level and the weight of each risk category.

9. The tender document review system according to claim 7, characterized in that, It also includes the MCP service layer, the A2A protocol layer, and the coordinator layer. The MCP service layer includes: The text conversion module is used to perform format detection on the original documents of the tender documents and bidding documents, and call the corresponding file parsing tool to parse the documents based on the detection results; the original document type includes at least one of PDF, DOCX, DOC, TXT, and rich text format; the parsing results are converted into structured text according to the logical hierarchy of documents, chapters, clauses, tables and metadata; OCR tools are used to recognize and convert text in scanned documents or images into searchable text. A document highlighting tool used to insert highlighted annotations at specified locations and output them in PDF / DOCX format; The fault-tolerance module is used to correct non-standard analysis results. The correction includes at least pre-cleaning, layer-by-layer regularization, and structured or manual review. The non-standard results include missing quotation marks, mismatched brackets, and mixed use of single quotation marks. The A2A protocol layer is used to define the message format and state machine for exchanging tasks between intelligent agents. The message format and state machine include JSON format for task requests / responses, status reports, error codes and retry mechanisms, as well as authentication and access control. The coordinator layer includes a coordinator that receives the original documents of tender documents and bidding documents. If the original documents contain scanned copies or images, the coordinator uses an OCR tool to recognize and convert the text in the scanned copies or images into searchable text. It performs format checks on the original documents of tender documents and bidding documents, and calls the corresponding file parsing tool to parse the documents based on the check results. It also uses an A2A protocol to schedule business compliance, technical compliance, and qualification matching agents to analyze the structured text of the tender documents to be reviewed, and calls an integration agent to integrate and optimize problem descriptions and modification suggestions. It calls a fault-tolerance module to correct non-standard analysis results and generate standardized analysis results. Finally, it calls a file highlighting tool to map the analysis results to the corresponding positions in the text of the tender documents to be reviewed according to paragraph index, page number, line number, or OCR coordinates, and highlights the corresponding positions to generate a tender document with highlighted annotations.

10. The tender document review system according to claim 7, characterized in that: Various intelligent agents are constructed using large-scale language models or local fine-tuning models, bidding documents, and various format and standard review prompts, and the intelligent agents communicate through the A2A protocol.

Images