Multi-agent collaborative risk control system and method for flexible employment

Abstract

This application discloses a multi-agent collaborative risk control system and method for flexible employment. The system is equipped with a supervisor agent and multiple worker agents. After system startup, it is used to: receive user input of risk control requirements for flexible employment scenarios; identify the intent of the risk control requirements; if the identified risk control requirements contain multiple intents, then combine the multiple intents to decompose the risk control requirements into multiple sub-tasks; based on a preset routing strategy, distribute each sub-task to the corresponding worker agent, and have each worker agent execute its sub-task, which includes policy compliance assessment, contract risk review, financial health assessment, consistency verification of four-stream data, root cause analysis of failures, and / or data statistics; monitor the task execution status of each worker agent through the supervisor agent, and after all sub-tasks are completed, integrate the output results of each worker agent to generate a flexible employment risk control report. This application can improve the efficiency and accuracy of flexible employment risk detection.

Description

Technical Field

[0001] This application relates to the fields of flexible employment and artificial intelligence technology, and in particular to a multi-agent collaborative risk control system and method for flexible employment. Background Technology

[0002] With the rapid development of the sharing economy, flexible employment has become an important form of employment in many industries due to its advantages of efficient and flexible resource allocation. To regulate the operation of flexible employment platforms, relevant industry standards clearly require platforms to have full-process service capabilities such as identity verification, income settlement, comprehensive risk control, and data statistics, covering the compliant management of relevant business data and personal information of various entities such as corporate users, individual users, and agents.

[0003] Meanwhile, the policy clarifies that the tax nature of the income of employees on the platform has been changed from the original assessment and collection of business income to the comprehensive cumulative withholding method for labor income. The platform is also required to fulfill legal obligations such as real-name verification, authorization retention, regular reporting, and summary declaration. This further strengthens the full-chain compliance requirements for the contract flow, business flow, capital flow, invoice flow, tax flow, and identity flow of flexible employment platforms.

[0004] However, existing flexible employment platforms' risk control systems suffer from the following problems: First, the risk control process involves collaboration across multiple professional departments, including customer relations, legal, finance, policy, and technology. Differences in staff professional competence, low communication efficiency, and staff turnover lead to unstable risk control decision-making quality. Second, the current risk control process is executed sequentially, taking weeks to months on average from risk data discovery and manual investigation to decision implementation. This makes it difficult to meet real-time monitoring requirements and may lead to secondary risks such as litigation and arbitration losses, and tax rating downgrades due to delayed responses. Third, flexible employment involves massive amounts of policy clauses, contract texts, and business data. A single manual or general-purpose system cannot fully cover the professional needs of multiple fields, easily leading to risk omissions. Furthermore, existing risk control systems are mostly single-function modules, lacking the ability to break down and collaboratively process complex risk control needs, making it difficult to handle complex risk control scenarios such as consistency verification of four flows, dynamic policy adaptation, and cross-system data linkage. Summary of the Invention

[0005] In view of this, embodiments of this application provide a multi-agent collaborative risk control system and method for flexible employment, which is used to solve at least one of the above-mentioned technical problems.

[0006] Firstly, embodiments of this application provide a multi-agent collaborative risk control system for flexible employment. The system is equipped with a supervisory agent and multiple working agents. The multiple working agents include at least two of the following: a policy expert agent, a legal and contract agent, a financial advisor agent, a project operation agent, a technical expert agent, and a data statistics agent. After the system is started, it is used to perform the following processes: Receive user input of risk control requirements for flexible employment scenarios. Risk control requirements include at least one of the following: compliance review, contract review, cost calculation, risk screening, and data statistics. Based on the fine-tuned semantic matching model, intent recognition is performed on risk control requirements; if multiple intents are identified in the risk control requirements, the risk control requirements are decomposed into multiple sub-tasks by combining the multiple intents. Based on a preset routing strategy, the decomposed sub-tasks are distributed to corresponding working agents, which then execute their respective sub-tasks. Among them, the policy expert agent is used to assess the policy compliance of the flexible employment project; the legal and contract agent is used to conduct risk review of the contract texts of the flexible employment project; the financial advisor agent is used to assess the financial health of the flexible employment project; the project operation agent is used to verify the consistency of the four-stream data of the flexible employment project; the technical expert agent is used to conduct root cause analysis of the flexible employment platform; and the data statistics agent is used to conduct data statistical analysis of the flexible employment project. The supervisory agent monitors the task execution status of each working agent, and after all sub-tasks are completed, the output results of each working agent are integrated to generate a flexible employment risk control report.

[0007] Secondly, embodiments of this application provide a multi-agent collaborative risk control method for flexible employment. This method is based on the multi-agent collaborative risk control system for flexible employment provided in the first aspect, and includes: Receive user input of risk control requirements for flexible employment scenarios. Risk control requirements include at least one of the following: compliance review, contract review, cost calculation, risk screening, and data statistics. Based on the fine-tuned semantic matching model, intent recognition is performed on risk control requirements; if multiple intents are identified in the risk control requirements, the risk control requirements are decomposed into multiple sub-tasks by combining the multiple intents. Based on a preset routing strategy, the decomposed sub-tasks are distributed to corresponding working agents, which then execute their respective sub-tasks. Among them, the policy expert agent is used to assess the policy compliance of the flexible employment project; the legal and contract agent is used to conduct risk review of the contract texts of the flexible employment project; the financial advisor agent is used to assess the financial health of the flexible employment project; the project operation agent is used to verify the consistency of the four-stream data of the flexible employment project; the technical expert agent is used to conduct root cause analysis of the flexible employment platform; and the data statistics agent is used to conduct data statistical analysis of the flexible employment project. The supervisory agent monitors the task execution status of each working agent, and after all sub-tasks are completed, the output results of each working agent are integrated to generate a flexible employment risk control report.

[0008] The multi-agent collaborative risk control system and method for flexible workforce provision in this application, on the one hand, achieves specialized division of risk control tasks through a collaborative architecture of a supervisory agent and multiple vertical domain working agents, effectively solving the efficiency bottleneck problem caused by traditional cross-departmental serial collaboration. Specialized agents such as policy experts, legal and contract / financial advisors perform their respective duties, while the supervisory agent monitors the entire process, enabling orderly scheduling and result integration of multi-agent collaborative work, improving risk control response timeliness, and reducing secondary risks such as litigation and arbitration losses and tax rating downgrades caused by delayed responses. On the other hand, through a fine-tuned semantic matching model and preset routing strategies, complex risk control needs can be accurately identified and broken down into dedicated sub-tasks, distributed to corresponding specialized agents for processing, avoiding the shortcomings of single manual or general-purpose models in covering multiple domain professional needs. Each working agent focuses on a specific domain, accurately completing professional tasks such as policy compliance assessment, four-stream data consistency verification, and root cause analysis of failures, effectively reducing risk omissions caused by differences in personnel professional competence and knowledge gaps, and improving the accuracy and stability of risk control decisions. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings of the embodiments of this application will be briefly described below.

[0010] Figure 1 This is a structural block diagram of a multi-agent collaborative risk control system for flexible workforce provided in an embodiment of this application.

[0011] Figure 2 This is a schematic diagram of the execution flow of a multi-agent collaborative risk control system for flexible employment provided in an embodiment of this application.

[0012] Figure 3 This is a schematic diagram of the execution flow of a policy expert intelligent agent provided in an embodiment of this application.

[0013] Figure 4 This is a schematic diagram of the execution flow of a legal and contract intelligent agent provided in an embodiment of this application.

[0014] Figure 5 This is a schematic diagram of the execution flow of a financial advisory intelligent agent provided in an embodiment of this application.

[0015] Figure 6 This is a schematic diagram of the execution flow of a project operation intelligent agent provided in an embodiment of this application.

[0016] Figure 7 This is a schematic diagram of the execution flow of a technical expert intelligent agent provided in an embodiment of this application.

[0017] Figure 8 This is a schematic diagram of the execution flow of a data statistics intelligent agent provided in an embodiment of this application.

[0018] Figure 9 This is a schematic diagram of an architecture for a multi-agent collaborative risk control system for flexible employment, provided in an embodiment of this application. Detailed Implementation

[0019] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0020] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0021] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0022] Various modifications and variations can be made to this application without departing from its spirit or scope, which will be apparent to those skilled in the art. Therefore, this application is intended to cover modifications and variations falling within the scope of the corresponding claims (the claimed technical solutions) and their equivalents. It should be noted that the implementation methods provided in the embodiments of this application can be combined with each other without contradiction.

[0023] Before describing the technical solutions provided in the embodiments of this application, in order to facilitate understanding of the embodiments of this application, this application first specifically explains the problems existing in the prior art: With the rapid development of the sharing economy, flexible employment has become an important form of employment in many industries due to its advantages of efficient and flexible resource allocation. To regulate the operation of flexible employment platforms, relevant industry standards clearly require platforms to have full-process service capabilities such as identity verification, income settlement, comprehensive risk control, and data statistics, covering the compliant management of relevant business data and personal information of various entities such as corporate users, individual users, and agents.

[0024] Meanwhile, the policy clarifies that the tax nature of the income of employees on the platform has been changed from the original assessment and collection of business income to the comprehensive cumulative withholding method for labor income. The platform is also required to fulfill legal obligations such as real-name verification, authorization retention, regular reporting, and summary declaration. This further strengthens the full-chain compliance requirements for the contract flow, business flow, capital flow, invoice flow, tax flow, and identity flow of flexible employment platforms.

[0025] However, existing flexible employment platforms' risk control systems suffer from the following problems: First, the risk control process involves collaboration across multiple professional departments, including customer relations, legal, finance, policy, and technology. Differences in staff professional competence, low communication efficiency, and staff turnover lead to unstable risk control decision-making quality. Second, the current risk control process is executed sequentially, taking weeks to months on average from risk data discovery and manual investigation to decision implementation. This makes it difficult to meet real-time monitoring requirements and may lead to secondary risks such as litigation and arbitration losses, and tax rating downgrades due to delayed responses. Third, flexible employment involves massive amounts of policy clauses, contract texts, and business data. A single manual or general-purpose system cannot fully cover the professional needs of multiple fields, easily leading to risk omissions. Furthermore, existing risk control systems are mostly single-function modules, lacking the ability to break down and collaboratively process complex risk control needs, making it difficult to handle complex risk control scenarios such as consistency verification of four flows, dynamic policy adaptation, and cross-system data linkage.

[0026] To address at least one of the aforementioned technical problems, this application provides a multi-agent collaborative risk control system and method for flexible employment.

[0027] The following section first introduces the multi-agent collaborative risk control system for flexible employment provided in this application.

[0028] Figure 1 This is a structural block diagram of a multi-agent collaborative risk control system for flexible workforce provided in an embodiment of this application. Figure 1 As shown in the embodiment of this application, the multi-agent collaborative risk control system 10 for flexible employment is equipped with a supervisor agent 101 and multiple working agents. The multiple working agents may include at least two of the following: a policy expert agent 102, a legal and contract agent 103, a financial advisor agent 104, a project operation agent 105, a technical expert agent 106, and a data statistics agent 107. Each working agent is configured with a workflow for handling specific tasks in flexible employment scenarios, and processes these specific tasks based on models, algorithms, and / or knowledge bases fine-tuned for flexible employment scenarios.

[0029] Among them, policy expert agent 102 can be used to assess the policy compliance of flexible employment projects. Legal and contractual agent 103 can be used to conduct risk review of contract texts for flexible employment projects. Financial advisor agent 104 can be used to assess the financial health of flexible employment projects. Project operation agent 105 can be used to verify the consistency of the four-stream data of flexible employment projects. Technical expert agent 106 can be used to conduct root cause analysis of failures on the flexible employment platform. Data statistics agent 107 can be used to conduct data statistical analysis of flexible employment projects.

[0030] Figure 2This is a schematic diagram of the execution flow of a multi-agent collaborative risk control system for flexible workforce provided in an embodiment of this application. Figure 2 As shown, the multi-agent collaborative risk control system for flexible employment (hereinafter referred to as the multi-agent collaborative risk control system) 10 can be used to perform the following steps S201 to S204 after startup.

[0031] S201: Receive user input of risk control requirements for flexible employment scenarios. Risk control requirements include at least one of the following: compliance review, contract review, cost calculation, risk screening, and data statistics.

[0032] In some embodiments, the multi-agent collaborative risk control system 10 can be a dedicated risk control subsystem embedded in a Flexible Employment Platform (FEP) system. Users can access the multi-agent collaborative risk control system 10 through the interactive portal of the Flexible Employment Platform system. For example, in some specific embodiments, users can click the button to access the multi-agent collaborative risk control system 10 on the project details page of the Flexible Employment Platform operation system, and input their flexible employment scenario risk control requirements in natural language through a dialog window, such as "Please conduct a comprehensive compliance and risk analysis on the data of the campus bicycle maintenance project over the past month, focusing on whether there are any individuals with inconsistent employment status or abnormal salary payments." The flexible employment scenario risk control requirements (hereinafter referred to as risk control requirements) can include at least one of the following: compliance review, contract review, cost calculation, risk screening, and data statistics. The multi-agent collaborative risk control system 10 can support the reception of single-dimensional or multi-dimensional composite risk control requirements.

[0033] S202: Based on the fine-tuned semantic matching model, perform intent recognition on risk control requirements; if the risk control requirements are found to contain multiple intents, then combine the multiple intents and decompose the risk control requirements into multiple sub-tasks.

[0034] Upon receiving a risk control request, the multi-agent collaborative risk control system 10 can invoke a fine-tuned semantic matching model to semantically parse the user's input risk control request, described in natural language, and identify the user's risk control intent. If the risk control request contains multiple intents, such as simultaneously requiring policy compliance review and four-stream data consistency verification, then based on these multiple intents and the specialized risk control functions of multiple working agents, the risk control request is broken down into multiple independent sub-tasks, each corresponding to a working agent. If the risk control request contains only one intent, then that intent is transformed into a corresponding specialized sub-task.

[0035] S203: Based on the preset routing strategy, the decomposed subtasks are distributed to the corresponding working agents, and each working agent executes the subtasks respectively.

[0036] The preset routing strategy can include a mapping relationship between risk control subtask types and working agents. Based on this mapping relationship, the decomposed subtasks can be distributed to the corresponding working agents.

[0037] Upon receiving a subtask, each working agent initiates a workflow to process the specific task and executes it. For example, policy expert agent 102 can retrieve the latest tax policies from the policy database and compare the compliance of flexible employment projects with the latest tax policies. Project operation agent 105 can retrieve four-stream data from systems such as Customer Relationship Management (CRM), Flexible Employment Platform (FEP), and / or Big Data Service (BDS) and perform consistency verification of the four-stream data. Data statistics agent 107 can perform data statistical analysis on flexible employment projects.

[0038] In some embodiments, risk control requirements may contain only one intent. If it is identified that a risk control requirement contains only one intent, such as data statistical analysis, the risk control requirement is transformed into a corresponding specific sub-task, and based on a preset routing strategy, the specific sub-task is distributed to the corresponding working agent, which then executes the specific sub-task.

[0039] S204: Monitor the task execution status of each working agent through the supervisor agent, and integrate the output results of each working agent after all sub-tasks are completed to generate a flexible employment risk control report.

[0040] The supervisor agent 101 performs the functions of the Project Management Office (PMO) team in risk control. It is mainly used to monitor the task execution status of each working agent, coordinate the collaboration logic between working agents, and integrate the output results of all working agents.

[0041] During the execution of sub-tasks by each working agent, the supervisor agent 101 can monitor the execution status of all sub-tasks in real time, such as whether the task has started, the execution progress, and whether there are any missing or abnormal data. Once all sub-tasks are completed, the supervisor agent 101 collects the output results from each working agent, such as policy compliance review opinions, four-stream consistency verification results, financial health assessment reports, and / or statistical reports. It then integrates these results according to a preset report structure template to generate a flexible employment risk control report. The multi-agent collaborative risk control system 10 supports online editing and multiple rounds of follow-up questions to optimize the flexible employment risk control report, ensuring that it meets the user's actual risk control decision-making needs.

[0042] The multi-agent collaborative risk control system for flexible staffing provided in this application, on the one hand, achieves specialized division of risk control tasks through a collaborative architecture of a supervisory agent and multiple vertical domain working agents, effectively solving the efficiency bottleneck problem caused by traditional cross-departmental serial collaboration. Specialized agents such as policy experts, legal and contract / financial advisors perform their respective duties, while the supervisory agent monitors the entire process, enabling orderly scheduling and result integration of multi-agent collaborative work. This improves the timeliness of risk control response and reduces secondary risks such as litigation and arbitration losses, and tax rating downgrades caused by delayed responses. On the other hand, through a fine-tuned semantic matching model and preset routing strategies, complex risk control needs can be accurately identified and broken down into dedicated sub-tasks, which are then distributed to corresponding specialized agents for processing. This avoids the shortcomings of single manual or general-purpose models in covering multiple domain professional needs. Each working agent focuses on a specific domain and can accurately complete professional tasks such as policy compliance assessment, four-stream data consistency verification, and root cause analysis of failures. This effectively reduces risk omissions caused by differences in personnel professional competence and knowledge gaps, improving the accuracy and stability of risk control decisions.

[0043] To facilitate understanding, the following examples illustrate a multi-agent collaborative risk control system for flexible workforce utilization.

[0044] Optionally, according to some embodiments of this application, the multi-agent collaborative risk control system 10 can also be used to perform the following steps: Before distributing the decomposed subtasks to the corresponding working agents, determine whether there are any dependencies between the subtasks.

[0045] Specifically, a dependency rule base can be pre-defined, which can contain multiple dependency rules. For example, the tax compliance assessment subtask might depend on the cash flow verification subtask, or the four-flow consistency verification subtask might depend on the contract review subtask. Based on the dependency rules in the dependency rule base, it can be determined whether there are dependencies between the various subtasks.

[0046] Accordingly, S203: Based on a preset routing strategy, the decomposed subtasks are distributed to their respective working agents, which may include the following steps: If there are no dependencies, each subtask will be distributed to the corresponding working agents for parallel execution based on the preset routing strategy. If a dependency exists, the working agent corresponding to each subtask is determined based on a preset routing strategy, and the execution order of each subtask is determined. The supervisor agent starts the corresponding working agents in sequence according to the execution order, and uses the output of the upstream working agent as the input of the downstream working agent.

[0047] Specifically, if there are dependencies between subtasks, the corresponding working agent for each subtask can be determined one by one based on a preset routing strategy. Secondly, based on the dependency analysis results, the execution order of each subtask is determined. The supervisor agent then triggers the corresponding working agents to execute the subtasks sequentially according to their execution order. During execution, the supervisor agent can synchronize the output of the upstream working agents to the downstream working agents as input. For example, after the upstream legal and contract agents complete the contract review subtask, they synchronize the results to the project operations agent, allowing the project operations agent to execute the four-flow consistency verification subtask.

[0048] Thus, for subtasks without dependencies, the routing strategy precisely matches the working agents and executes them in parallel, reducing the efficiency waste of the traditional serial mode and shortening the risk control cycle. Furthermore, through the collaboration of routing strategies, dependencies, and the supervisor agent, the system can flexibly respond to the complex and diverse risk control needs in flexible employment scenarios. Whether it's an independent single subtask, a combination of multiple dependent subtasks, or a complex task chain with complex dependencies, it can achieve orderly and efficient processing, improving the system's adaptability to different risk control scenarios.

[0049] According to some embodiments of this application, optionally, risk control requirements may include a scenario description of flexible employment projects. Flexible employment projects include, but are not limited to, various projects applicable to flexible employment, such as campus shared bicycle maintenance services, remote home copywriting services, and on-site appliance repair services. The scenario description of a flexible employment project may include information such as service scenario, service method, delivery form, settlement method, and / or personnel scale. For example, service scenarios may include campuses, homes, commercial complexes, or outdoor venues. Service methods may include on-site service, remote service, platform-assigned orders, or self-employed orders. Delivery forms may include physical delivery, digital deliverables, delivery of uploaded vouchers (such as photo / video retention), or service acceptance confirmation. Settlement methods may include daily, weekly, monthly, or per-order settlement. Personnel scale may include quantitative ranges such as less than 10 people, 10-50 people, 50-200 people, or more than 200 people; this application does not limit this range.

[0050] Figure 3 This is a schematic diagram of the execution flow of a policy expert intelligent agent provided in an embodiment of this application. Figure 3 As shown, the policy expert agent can be used to perform the following steps S301 to S305.

[0051] S301: Extract key feature words from the scene description and construct a multi-dimensional scene profile based on the extracted key feature words. The multi-dimensional profile may include at least two of the following: service scene, service method, delivery form, settlement method, and personnel scale.

[0052] The policy expert agent can preprocess the scenario description, removing redundant modifiers that have no practical meaning. Then, it extracts key feature words from the preprocessed scenario description. For example, for the scenario description "Home-based remote data annotation service, flexible order taking, daily payment, approximately 80 employees, delivery format is annotated structured data", key feature words such as "home-based remote", "data annotation", "flexible order taking", "daily payment", "80 employees", and "structured data delivery" can be extracted.

[0053] Based on the extracted key features, at least two of the following are selected as core dimensions: service scenario, service method, delivery form, settlement method, and personnel scale, to construct a multi-dimensional scenario profile. For example, taking the above scenario as an example, the constructed scenario profile could be: service scenario is home-based, service method is remote, settlement method is daily settlement, personnel scale is 80 people, and delivery form is structured data.

[0054] S302: Based on scenario profiling, select candidate policies applicable to flexible employment projects from the policy database, and analyze the applicable scenario characteristics, core compliance requirements and tax incentive conditions of each candidate policy to generate policy feature tags.

[0055] The policy expert agent uses scenario profiles as filtering criteria to match candidate policies applicable to the current flexible employment project from a pre-set policy database. Optionally, in some embodiments, candidate policies selected based on scenario profiles can be further filtered based on region and / or industry to obtain the final candidate policies. For example, for the aforementioned home-based remote data annotation project, candidate policies A, B, and C are selected.

[0056] Then, the unstructured text of each candidate policy is decomposed into structured data containing applicable scenario characteristics, core compliance requirements, and tax incentive conditions, and corresponding policy feature labels are generated. For example, after parsing candidate policy A, the following policy feature labels are generated: applicable scenario characteristics are remote flexible employment and free order taking mode; core compliance requirements are real-name verification, cumulative withholding declaration of labor income, and retention of task data; tax incentive conditions are that those below the individual income tax threshold can enjoy tax reductions and exemptions.

[0057] S303: Calculate the similarity between the scenario profile and the policy feature labels of each candidate policy, and determine the target matching policy based on the similarity matching results.

[0058] In S303, the similarity between the scene profile and the policy feature labels of each candidate policy can be calculated, such as cosine similarity. The similarity score quantifies the degree of fit between the scene and the candidate policy. For example, the scene profile of the aforementioned home remote data annotation project has a similarity score of 93% with the policy feature labels of candidate policy A, 89% with the policy feature labels of candidate policy B, and 35% with the policy feature labels of candidate policy C.

[0059] Based on the similarity scores, candidates are sorted from highest to lowest, and at least one of the top-ranked policies is selected as the target matching policy. Key attributes such as the document number, effective date, and applicable region of each target matching policy are then labeled to ensure traceability of the basis for subsequent compliance determination. In the example above, candidate policy A was ultimately determined as the target matching policy.

[0060] S304: Based on a well-trained policy interpretation model, interpret the target matching policy, and combine the scenario profile and policy interpretation results to determine the compliance of flexible employment projects.

[0061] For example, a dedicated policy interpretation model can be fine-tuned from a general model using policy texts and historical cases related to flexible employment projects. In S304, target-matching policies and scenario profiles of flexible employment projects can be input into the dedicated policy interpretation model. This model can then interpret the target-matching policies, producing interpretation results. These results are then compared and verified against the scenario profiles of flexible employment projects to determine whether the projects meet the applicable scenario requirements and core compliance requirements of the target-matching policies.

[0062] If the scenario profile of a flexible employment project meets the applicable scenario requirements and core compliance requirements of the target matching policy, the flexible employment project is deemed compliant (encouraged). If the scenario profile meets the applicable scenario requirements and fulfills the mandatory core compliance requirements, but only lacks non-core compliance requirements such as filing and material submission, the flexible employment project is deemed compliant (under improvement). If the scenario profile does not meet the applicable scenario requirements, fails to meet the mandatory core compliance requirements, or violates policy prohibitions, the flexible employment project is deemed non-compliant.

[0063] S305: If the flexible employment project is determined to be compliant, then match the tax relief policy catalog, verify the eligibility for tax incentives, sort out the policy basis, and output the policy compliance assessment results.

[0064] If a flexible employment project is deemed compliant, the policy expert agent further matches it with a list of tax relief policies to verify whether the project qualifies for tax incentives. For example, for the aforementioned home-based remote data labeling project, it matches two preferential policies: exemption from individual income tax for monthly labor income of flexible workers below 5,000 yuan and exemption from value-added tax for monthly sales of small-scale taxpayers below 100,000 yuan. Considering the project's actual situation of "daily payment of approximately 150 yuan and average monthly income of 4,500 yuan," it is determined that the project qualifies for both tax incentives.

[0065] Next, the policy basis is reviewed, clarifying the names, document numbers, specific tax reduction and exemption ratios, application procedures, and required materials for each preferential policy. Finally, the compliance assessment conclusions, the list of target-matching policies, the results of the tax incentive eligibility assessment, and the policy basis are integrated according to a pre-set template to output a complete policy compliance assessment result.

[0066] In this way, by providing a simplified interpretation of the policy framework and verifying compliance requirements item by item, compliance risks such as tax penalties and platform rating downgrades caused by misunderstandings of policies are reduced. Furthermore, automatically verifying eligibility for tax incentives and streamlining the application process helps platforms and employees enjoy tax benefits, effectively reducing the costs associated with flexible employment.

[0067] According to some embodiments of this application, optionally, the policy compliance assessment results may include at least one of the following: compliance determination conclusion, target-matching policy list, tax incentive eligibility determination results, compliance rectification suggestions, and policy risk warning alerts. Policy risk warning alerts may include policy timeliness risks, regional adaptability risks, clause change risks, and / or industry application restriction risks. For example, risks such as a policy expiring soon, a policy only applicable to specific provinces / cities, a notice of adjustment to core policy clauses, or policies imposing employment restrictions on high-risk operations / special industries.

[0068] Figure 4 This is a schematic diagram illustrating the execution flow of a legal and contract intelligent agent provided in an embodiment of this application. Figure 4 As shown, according to some embodiments of this application, optionally, the legal and contract intelligent agent is specifically used to perform the following steps S401 to S405.

[0069] S401: When risk control requirements include the intent to review contracts, obtain the contract text for flexible employment projects.

[0070] When risk control needs include contract compliance review and clause risk screening, the legal and contract intelligence agents can retrieve the contract text corresponding to the flexible employment project from the business database of the flexible employment platform, or directly receive the contract text of the flexible employment project uploaded by the user.

[0071] S402: Call the contract text structured parsing model to split the contract document into at least one of the following modules: contract body module, service content module, payment terms module, liability for breach of contract module, data compliance module, confidentiality obligation module, and contract termination module, and semantically encode the content of the terms in each module to obtain a set of contract terms.

[0072] The contract text structured parsing model can be a model fine-tuned from a general large model using contract samples from the flexible employment industry. In S402, the contract text structured parsing model is invoked, splitting the contract document into at least one of the following modules: contract body module, service content module, payment terms module, breach of contract liability module, data compliance module, confidentiality obligation module, and contract termination module. Then, the clause content within each module is standardized with semantic encoding. That is, unstructured natural language clauses are transformed into machine-readable vector forms, while retaining basic information such as the original text and clause location. Finally, the semantic encoding results and original text information of all modules are integrated to obtain the contract clause set.

[0073] For example, taking the breakdown of a home-based remote copywriting project cooperation contract as an example, the information of the cooperating party in the contract can be classified into the contract body module, the copywriting delivery information into the service content module, and the settlement information and breach of contract information into the payment terms module. The terms of each module are then semantically encoded to obtain a set of contract terms.

[0074] S403: Compare the set of contract terms with the standard contract template in the contract resource library to identify any missing, inconsistent, or conflicting terms in the set of contract terms.

[0075] The contract resource library can contain multiple standard contract templates corresponding to different service types or scenarios. In S403, the set of contract terms can be compared with the standard contract template for the service type or scenario to which the contract belongs, identifying missing terms in the set, such as missing clauses regarding the division of safety responsibilities. In some examples, semantic similarity calculations can also be used to find whether there are clauses in the set that are inconsistent with and / or conflict with the wording of the standard contract template.

[0076] After the identification is completed, the issues in the identified contract clause set can be classified and labeled, including the issue type, the corresponding standard contract template clauses, and the deviation content, to obtain the contract standardization compliance review results.

[0077] For example, by comparing the contract terms of a commercial complex equipment inspection project contract with the corresponding standard contract template, it can be identified that the breach of contract liability module in the contract terms is missing the clause on the division of liability for equipment damage, and the settlement cycle in the payment terms module is 3 months, which is inconsistent with the monthly settlement statement in the standard contract template. These two types of problems are marked.

[0078] S404: Match the set of contract terms with legal provisions in a legal knowledge base to identify clauses in the set of contract terms that violate mandatory legal provisions and / or have potential legal loopholes.

[0079] S404 allows for the matching of contract terms against legal provisions in a legal knowledge base to identify clauses that violate mandatory legal provisions. For example, a contract stipulating that the service provider is not liable for work-related injuries violates relevant laws. It also identifies clauses that lack clear responsibilities or contain potential legal loopholes, such as the lack of a clause specifying the final payment method upon contract termination. Furthermore, it verifies whether the terms comply with regulatory requirements, such as whether cross-regional service contracts clearly define local registration responsibilities.

[0080] After identification, all legally problematic clauses in the contract clause set can be marked, such as the basis for violation, the type of loophole, and legal risks, to obtain the contract legal compliance review results.

[0081] S405: Based on the compliance review results, generate and output the contract risk audit results.

[0082] The standardization compliance issues of S403 and the legal compliance issues of S404 are summarized and categorized according to contract module dimensions (such as contract subject, service content, payment terms, liability for breach of contract, etc.). Standardization deviation issues and legal issues within the same module are presented together. Simultaneously, a mapping system is established between problematic clauses and corresponding clauses in the original contract text, standard contract templates, and relevant legal provisions. This ensures that each issue can be traced back to the specific clause location, comparison basis, and legal basis, resulting in a contract risk assessment.

[0083] Thus, when risk control requirements include contract review, the entire process of flexible employment contract review is automated through legal and contract intelligence, replacing the traditional manual, word-by-word review model. This improves the efficiency of contract risk review and effectively shortens the overall risk control cycle for flexible employment projects. On the other hand, by constructing a dual verification system of standard contract templates and a legal knowledge base, contract clause issues are comprehensively investigated from both standardization and legal compliance dimensions. This avoids subjective biases and omissions in manual review, improving the accuracy and comprehensiveness of contract risk identification and reducing potential risks at the contract clause level from the source.

[0084] According to some embodiments of this application, the risk control requirements may optionally include disputes or inquiries from users regarding contract documents. For example, disputes concerning the determination of payment milestones during contract performance, inquiries regarding discrepancies between service content and contract stipulations, inquiries related to the determination of liability for breach of contract, and disputes concerning the final payment settlement after contract termination.

[0085] Accordingly, legal and contractual intelligent agents can also be used to execute steps one through three.

[0086] Step 1: Identify the intent behind the dispute or consultation and extract the core demands.

[0087] Semantic analysis and intent recognition are performed on user disputes or inquiries regarding contract documents to remove redundant expressions and identify the user's core demands.

[0088] For example, if a user submits an inquiry titled "A partner company failed to pay the service fee by the 5th of each month as stipulated in the contract, and then refused to pay the penalty for late payment, what should be done?" After intent identification, the user's core demand is extracted as a claim for breach of contract liability for late payment and a solution.

[0089] Step Two: Analyze the disputed or consulted matters by combining contract terms, legal provisions, and user performance history data, and generate a dispute resolution solution; the dispute resolution solution includes at least one of the following: a negotiation plan, legal remedies, and a list of evidence.

[0090] The system identifies contract clauses relevant to the user's core needs from a set of contract terms, such as payment terms and breach of contract clauses. Simultaneously, it matches corresponding legal provisions from a legal knowledge base and retrieves historical performance data of both parties from the business system. For example, historical performance data may include payment records, breach of contract records, and contract performance status.

[0091] Based on contract terms, legal provisions, and user performance history data, the legal and contract intelligence agent comprehensively analyzes disputes or inquiries, and generates targeted dispute solutions by combining industry practices in flexible employment scenarios. These solutions include at least one of the following: a negotiation plan, legal remedies, and a list of evidence. For example, by combining a clause in the contract stating that "late payments will incur a penalty of 0.05% per day," relevant legal provisions on breach of contract, and the company's performance history of three instances of late payment, and referencing the industry practice of negotiating before pursuing liability for overdue payments, the legal and contract intelligence agent can generate a solution that includes negotiation scripts, arbitration remedies, and settlement statements as evidence.

[0092] Step 3: Integrate dispute resolution methods into the contract risk review results.

[0093] Following a pre-defined structured format, dispute resolution solutions are integrated into the contract risk audit results. For example, in some examples, a dispute or consultation handling module can be added to the contract risk audit results, presenting the focus of the dispute, the core analytical basis, and specific solutions. This integrates the results of contract risk audit and dispute consultation handling, allowing users to conveniently view contract audit results and dispute handling guidelines in one place.

[0094] In this way, the legal and contract intelligence agent extends its capabilities from simple contract risk assessment to professional handling of contract disputes or consultations, meeting users' risk control and legal needs throughout the entire contract lifecycle. Furthermore, integrating dispute resolution solutions with contract risk assessment results enables a fusion of these two types of outcomes, providing users with a one-stop reference for contract risk control and dispute resolution, thus improving user experience and decision-making efficiency.

[0095] Figure 5 This is a schematic diagram of the execution flow of a financial advisory intelligent agent provided in an embodiment of this application. Figure 5 As shown, according to some embodiments of this application, optionally, the financial advisor agent is specifically used to perform the following steps S501 to S506.

[0096] S501: When risk control requirements include the intent of financial health assessment, obtain basic data and cost calculation parameters for flexible staffing projects.

[0097] When risk control requirements include the intent of financial health assessment, the financial advisor agent can retrieve basic data and cost calculation parameters of flexible employment projects from the business database of the flexible employment platform based on the project's identifier.

[0098] The basic data for flexible staffing projects may include project initiation information, service type, service scale, partner entity information, expected service period, historical data on similar projects, and / or employee salary standards. Cost calculation parameters for flexible staffing projects may include corporate financing costs, market interest rates, industry risk coefficients, personnel costs, operating and management costs, platform service fee ratio, advance payment amount, advance payment period, financing interest rate, revenue forecast base, tax accrual ratio, payment channel fees, and / or technical service costs.

[0099] For example, to address the risk control needs of a campus shared bicycle operation and maintenance project's financial health assessment, the financial advisor's intelligent agent retrieves the project's basic data, such as a 1-year service period, coverage of 3 universities, 50 operation and maintenance personnel, and an average monthly salary of 4,000 yuan per person. It also retrieves the project's cost calculation parameters, such as a 5% platform service fee, an estimated advance payment of 200,000 yuan, an advance payment period of 3 months, a monthly financing interest rate of 0.5%, and an estimated monthly revenue of 150,000 yuan.

[0100] S502: Verify, standardize, and complete missing parameters of the basic data and cost calculation parameters to obtain the project calculation parameters.

[0101] First, compare the data with the cost assessment template to check the completeness of the basic data and cost calculation parameters. For example, check for missing key parameters such as information on partners and the funding period. Also, remove outliers and errors, such as parameters with salary standards exceeding the industry average. Second, standardize data in different formats and with different definitions, unifying data units, statistical dimensions, and calculation methods. For example, convert the service fee percentages of different partners to an annualized rate. Third, use interpolation, historical data from similar projects, or industry averages to complete any missing parameters and obtain the project calculation parameters.

[0102] S503: Based on industry benchmark data obtained from the business system, the project calculation parameters are calibrated to obtain the calibrated project calculation parameters.

[0103] Retrieve industry benchmark data for the corresponding service type in the flexible employment sector from the business system, such as average labor costs, average operating costs, and benchmark ROI. Based on this industry benchmark data, perform multi-dimensional calibration on the project calculation parameters to obtain the calibrated project calculation parameters.

[0104] For example, industry benchmark data for flexible staffing projects in campus operations and maintenance can be retrieved, such as the industry average labor cost per employee being 4,500 yuan / month, the average operating cost as a percentage of revenue being 12%, and the benchmark return on investment being 10%. Based on the industry benchmark data, the project calculation parameters are calibrated. For example, the initial calculated labor cost per employee of 5,500 yuan / month (22% higher than the benchmark) is lowered to 4,600 yuan / month, and the operating cost as a percentage of revenue is optimized from 18% (50% higher than the benchmark) to 13%, resulting in the calibrated project calculation parameters.

[0105] S504: Based on the calibrated project calculation parameters, at least two of the trained net present value calculation model, market capitalization-to-revenue ratio valuation model, and cost-profit ratio prediction model are called in parallel to calculate the preliminary financial indicators of the flexible employment project.

[0106] Based on the calibrated project calculation parameters, at least two of the trained net present value calculation model, market capitalization-to-revenue ratio valuation model, and cost-profit ratio prediction model can be called in parallel computing mode to perform synchronous financial calculations.

[0107] For example, the net present value (NPV) calculation model can be used to determine the project discount rate by calling the weighted average cost of capital (WACC) calculation algorithm, combined with corporate financing costs, market interest rates, and industry risk coefficients; it can also break down the project's annual revenue with annual costs such as labor costs, technology costs, and advance payment interest, and substitute these into the NPV formula to calculate the sum of the present values ​​of net cash flows for each year within the cooperation period, thus obtaining financial indicators such as the project's NPV. The market capitalization-to-revenue ratio valuation model can be used to obtain a benchmark value for the market capitalization-to-revenue ratio in the flexible employment industry from industry data; based on the project's annual projected revenue and incorporating customer credit rating adjustment factors, it can calculate the project's valuation, and then compare it with the project's estimated total investment costs to calculate financial indicators such as the project's valuation return rate. The cost-profit margin prediction model can be used to predict financial indicators such as the project's annual cost details, annual revenue, and annual profit margin based on factors such as personnel size, service fee rates, and advance payment periods, and by calling historical project training data.

[0108] Each model can operate independently without interfering with others, which can shorten the overall calculation time. After each model outputs the corresponding financial indicator calculation results, the financial advisor AI integrates all the results to obtain the preliminary calculation results of the financial indicators of the flexible employment project.

[0109] S505: Conduct sensitivity analysis on key parameters affecting project profits, and simulate parameter fluctuations using the Monte Carlo simulation algorithm to obtain sensitivity analysis results, including profit fluctuation ranges and risk warning points.

[0110] The financial advisor's intelligent agent first conducts sensitivity analysis on key parameters affecting project profits (such as revenue forecasts, labor costs, advance payments, and / or financing rates), calculating the impact of fluctuations in each key parameter on project profits and identifying profit-sensitive core parameters with a significant impact on project profits. Then, using a Monte Carlo simulation algorithm, multiple random sampling simulations are performed on the identified profit-sensitive core parameters to simulate random fluctuations within a reasonable range, and the project profit results under different fluctuation scenarios are statistically analyzed. Finally, the sensitivity analysis and simulation results are integrated to obtain the sensitivity analysis results. The sensitivity analysis results can include the project profit fluctuation range, profit threshold, and risk warning points corresponding to each profit-sensitive core parameter, such as a negative project profit when the financing rate exceeds 0.8%.

[0111] For example, sensitivity analysis of flexible employment projects in campus operation and maintenance revealed that revenue forecast and labor cost are the core sensitive parameters for profit. Monte Carlo simulation showed that the annualized fluctuation range of project profit is 50,000 to 120,000 yuan. At the same time, risk warning points were identified: when revenue is lower than 120,000 yuan / month or labor cost is higher than 2.2 million yuan / year, the project's return on investment will be lower than the industry benchmark by 8%.

[0112] S506: Based on the results of sensitivity analysis, actual financial data of similar historical projects, and customer performance records, cross-validate the preliminary calculation results to generate an assessment of the financial health status of the flexible employment project.

[0113] Based on the sensitivity analysis results of S505, the financial advisory agent simultaneously retrieves actual financial data from similar historical projects (such as the actual return on investment and cost-profit ratio of similar campus operation and maintenance projects) and the customer performance records of project partners (such as historical payment performance rate and cash flow capacity). It cross-validates the preliminary calculation results from four dimensions: model calculation, risk analysis, historical reference, and customer performance. Values ​​with significant deviations from actual conditions are removed from the preliminary calculations, and the results are revised. For example, if the actual profit rate of similar historical projects is 3% lower than the model prediction, the current predicted profit rate is revised to 12%. Based on the revised preliminary calculation results, an assessment of the financial health of the flexible employment project is generated.

[0114] Thus, the entire process of financial health assessment for flexible employment projects is automated through the intelligent financial advisory agent. From data acquisition, model calculation, risk analysis to result generation, it replaces manual calculation, shortens the financial assessment time, and improves the efficiency of project financial risk control. Secondly, a full-link financial calculation system is constructed, encompassing data preprocessing, dual parameter calibration, parallel calculation of multiple models, and multi-dimensional cross-validation. This system controls data and calculation accuracy at each level, effectively avoiding subjective bias and data errors inherent in manual calculations, and improving the accuracy and reliability of financial indicator calculation results. Furthermore, the introduction of sensitivity analysis and Monte Carlo simulation algorithms can accurately identify core profit-sensitive parameters, quantify the profit risk caused by parameter fluctuations, determine profit fluctuation ranges and risk warning points, and achieve dynamic risk analysis in financial assessment, thereby enhancing the ability to predict project financial risks.

[0115] According to some embodiments of this application, optionally, S503: Based on industry benchmark data obtained from the business system, the project calculation parameters are calibrated to obtain calibrated project calculation parameters, which may include the following steps four and five.

[0116] Step 4: Compare the key operating parameters in the project's calculation parameters with similar industry benchmark data, and based on the comparison results, generate calibration suggestions for key operating parameters that deviate from similar industry benchmark data.

[0117] In step four, key operational parameters are extracted from the project's calculation parameters, such as personnel labor costs, platform service fee ratio, and operational management cost ratio. These key operational parameters are then compared one by one with industry benchmark data for the same service type to calculate the parameter deviation. For key operational parameters whose deviation exceeds a preset threshold, targeted calibration suggestions are generated. For example, calibration suggestions may include calibration direction, calibration range, and reference basis.

[0118] For example, the annualized labor cost for flexible staffing projects in campus operations and maintenance is 2.4 million yuan, while the industry benchmark is 2 million yuan, a deviation of +20%. The financial advisor's intelligent agent generation calibration suggestion is as follows: the labor cost parameter is too high. It is recommended to optimize personnel allocation in conjunction with the industry benchmark and reduce the labor cost to the range of 2 million to 2.2 million yuan, based on the average labor cost standard for flexible staffing in the campus operations and maintenance industry.

[0119] Step 6: Update the project calculation parameters based on the calibration recommendations to obtain the calibrated project calculation parameters.

[0120] Based on the calibration recommendations, the key operational parameters in the project calculation parameters are updated to obtain the calibrated project calculation parameters. For example, the financial advisory agent adjusts the labor cost of a flexible employment project for campus operation and maintenance to 2.1 million yuan according to the calibration recommendations, while retaining the original advance payment interest parameter. The other parameters remain unchanged, and the updated parameters are the calibrated project calculation parameters.

[0121] In this way, by comparing key operating parameters with industry benchmark data, unreasonable parameters that deviate from industry levels can be identified and corrected in a timely manner, ensuring the objectivity and rationality of the calculation parameters and laying an accurate data foundation for subsequent financial calculations.

[0122] According to some embodiments of this application, optionally, the assessment results of the financial health status may include the calculation results of multiple financial indicators, a cost structure analysis table, and a conclusion on the financial feasibility of the project. These multiple financial indicators may include at least one of the following: project net present value, market capitalization-to-revenue ratio valuation, return on investment, static or dynamic payback period, and a projected cost-profit breakdown for 1-3 years.

[0123] For example, a cost structure analysis table can present the composition, amount, percentage, and comparison with industry benchmarks of various project costs. The conclusion of a project's financial feasibility assessment can be categorized as either financially feasible or not financially feasible.

[0124] For example, taking flexible employment projects related to campus operations and maintenance as an example, the assessment results of financial health status can include the calculation results of financial indicators, such as net present value of 92,000 yuan, annualized return on investment of 11.8%, dynamic investment payback period of 8.5 months, and detailed cost and profit forecasts for the next 3 years. It can also include a cost structure analysis table, such as labor costs accounting for 70%, operating costs accounting for 15%, service fees accounting for 5%, taxes accounting for 6%, and other costs accounting for 4%. It can also include a conclusion on financial feasibility. If it is financially feasible, the core basis is that the project's net present value is positive, the return on investment is higher than the industry benchmark by 8%, and the dynamic investment payback period is shorter than the project's service cycle.

[0125] Thus, the assessment of financial health includes multi-dimensional financial indicator calculations, covering core dimensions such as project value, return on investment, payback period, and cost-profit forecasts. This provides a comprehensive view of the project's financial status and can meet users' diverse financial analysis needs. Furthermore, by providing a clear conclusion on the project's financial feasibility and highlighting the core evidence, it helps users quickly determine whether the project is financially feasible, significantly improving the efficiency of risk control decisions.

[0126] Figure 6 This is a schematic diagram of the execution flow of a project operation intelligent agent provided in an embodiment of this application. Figure 6As shown, according to some embodiments of this application, optionally, risk control requirements may include the identifier of the flexible employment project and the time frame for risk screening. For example, the identifier of the flexible employment project may include the name and / or ID of the flexible employment project. The time frame for risk screening may be, for example, August to September 2025, and this application does not limit this.

[0127] Accordingly, the project operation intelligent agent is specifically used to perform the following steps S601 to S606.

[0128] S601: Based on the identification of flexible employment projects and the time range of risk investigation, retrieve the four-flow data related to the flexible employment project from multiple business systems. The four-flow data includes contract flow data, business flow data, fund flow data and invoice flow data.

[0129] The business systems include, but are not limited to, FEP systems, CRM systems, and / or BDS systems. Upon receiving a risk control request containing the identifier of the flexible employment project and the risk assessment timeframe, the project operation intelligent agent, based on the flexible employment project identifier, retrieves four-flow data from multiple flexible employment business systems within that risk assessment timeframe. Specifically, contract flow data may include project cooperation contracts, personnel service agreements, and other terms and performance records; business flow data may include task assignment vouchers, execution vouchers, acceptance vouchers, and completion vouchers; fund flow data may include service fee settlements, payroll payments, and bank transfer records; and invoice flow data may include invoice information, invoice authentication, and tax item matching data.

[0130] S602: Using personnel identifier and task identifier as the primary keys, construct data packets corresponding to each personnel based on the four-stream data.

[0131] The project operation intelligent agent can extract unified personnel identifiers (such as service personnel employee numbers and / or ID card numbers) and task identifiers (such as task numbers) from the four-flow data as association primary keys. Following the principle of "one person, one package; associated tasks within the package," it classifies and integrates cross-system and cross-type four-flow data. Specifically, it can aggregate contract flow, business flow, capital flow, and invoice flow data corresponding to all associated tasks of the same person within the risk assessment timeframe into the same data package, achieving precise association of four-flow data for a single person and a single task.

[0132] S603: Perform four-stream pass-through correlation verification on each data packet.

[0133] For each person's data packet, the project operation intelligent agent can perform four-flow penetration correlation verification, verifying the consistency of the four-flow data one by one according to the business logic of project operation.

[0134] S604: Verify the authenticity of business transactions from multiple dimensions. Multi-dimensional verification includes cross-validating multimodal task completion credentials in business flow data, verifying the consistency of personnel identity information at each stage, and verifying the rationality of business data based on historical business baseline data.

[0135] Building upon the four-flow penetration-based correlation verification, the project operation intelligent agent can perform multi-dimensional verification of the business authenticity of flexible employment projects, ensuring that business transactions actually occur and there are no fraudulent transactions. For example, it can cross-verify multimodal task completion vouchers (such as photos, videos, acceptance forms, electronic signature receipts, etc.) in the business flow data to verify the consistency of information across different modalities and eliminate the risk of forgery of a single voucher. Secondly, it can verify the consistency of personnel identity information at each stage of the four flows, ensuring that the identity of service personnel in the contract flow, the identity of task executors in the business flow, the identity of payees in the fund flow, and the identity information related to invoicing in the invoice flow are completely matched, preventing issues such as impersonation and fraudulent salary collection. Thirdly, it can retrieve historical business baseline data of similar projects within the platform, such as average task completion volume per person, single task time, and service unit price range, and compare the business data of flexible employment projects with historical business baseline data to verify the rationality of the business data and eliminate anomalies such as task volume, completion efficiency, and unit price deviating from reasonable ranges.

[0136] S605: Based on the results of correlation verification and business authenticity verification, locate the specific subject and link of abnormal data, and assess the risk level and impact scope of the abnormality according to preset rules.

[0137] For abnormal data in the four-stream penetration verification results and business authenticity verification results, the specific entity corresponding to the abnormal data is identified, such as a service personnel or a partner company, and the specific operational link corresponding to the abnormal data is identified, such as the task execution link, fund settlement link, and invoicing link. Then, according to the platform's preset risk assessment rules, the risk level of each abnormal data is assessed, and the impact of the abnormal data on the overall project operation, fund security, tax compliance, etc., is analyzed, such as involving only a single person or a single task, or affecting the entire project or multiple partners, to obtain the abnormal data risk assessment result. The risk assessment rules can be flexibly adjusted according to the actual situation, and this application does not limit them. For example, in some examples, abnormal data can be divided into different risk levels, such as low, medium, and high risk levels, according to the type and / or degree of deviation.

[0138] S606: Generate a consistency verification report for the four flows, including a list of abnormal data, risk level classification, and handling guidelines.

[0139] The project operation intelligent agent can structurally integrate abnormal data lists, risk level classifications, and handling guidelines to generate a standardized four-flow consistency verification report. For example, the four-flow consistency verification report may include an abnormal data list, risk level classifications, and handling guidelines, along with core evidence for four-flow penetration-based correlation verification and business authenticity verification, ensuring the report content is clear and well-supported. The handling guidelines may include suggestions on handling different risk levels and types of abnormal data, such as verification directions, rectification requirements, and handling timelines.

[0140] Thus, based on project identifiers and time ranges, precise retrieval of cross-system four-stream data is achieved. Combined with dual primary keys for personnel and tasks, a dedicated data package is constructed, enabling refined aggregation of four-stream data and resolving the issues of scattered data and difficulty in correlation across systems. Secondly, through four-stream penetration-based correlation verification and multi-dimensional business authenticity verification, risk assessment is conducted from both formal matching and substantive authenticity perspectives. This verifies the surface consistency of the four-stream data while confirming the actual occurrence of business transactions through multimodal credentials, identity verification, and baseline comparison, improving the comprehensiveness and accuracy of abnormal data identification. It can accurately identify compliance risks such as fraudulent transactions, false invoicing, and impersonation for salary payments. Furthermore, the generated four-stream consistency verification report, including anomaly lists, risk levels, and handling guidelines, provides clear handling suggestions for anomaly rectification, while achieving full-process traceability of the risk assessment process and results, meeting the dual needs of platform compliance operation and regulatory verification.

[0141] In some specific embodiments, optionally, S603: Perform four-stream pass-through correlation verification on each data packet, including at least one of the following: Perform correlation verification between contract flow data and business flow data to verify whether the task recipient has a valid contract and whether the task scope is consistent, and verify whether the task dispatch time is within the contract validity period. Perform correlation verification on business flow data and fund flow data, verify whether the settlement amount matches the task results, verify whether the payment time is later than the task acceptance time, and verify whether the identity of the payee is consistent. Perform correlation verification between fund flow data and invoice flow data, verify whether the invoice information is consistent with the payment amount and the payment entity, and verify whether the invoice issuance sequence conforms to the agreement; Build a complete data link from contract signing to invoice issuance, and identify whether there are missing or disconnected data gaps in the entire data link.

[0142] Specifically, when verifying the correlation between contract flow data and business flow data, the first step is to check whether the task recipient in the business flow is consistent with the contracting party of the valid contract in the contract flow, and whether the contract is in a normal effective state (e.g., not expired or terminated), ensuring that the task recipient has the legal qualifications to perform the contract. Secondly, it verifies whether the service scope in the business flow is within the service boundaries stipulated in the corresponding contract, identifying any non-compliant business exceeding the contract scope. Thirdly, it verifies whether the task assignment time and task execution cycle in the business flow are within the validity period stipulated in the contract, avoiding compliance risks associated with executing related tasks after the contract has expired.

[0143] When verifying the correlation between business flow data and cash flow data, the first step is to calculate the settlement amount based on the task completion volume, acceptance results, and agreed unit price in the business flow. This settlement amount is then compared with the actual payment amount or transfer amount in the cash flow to verify consistency and identify issues such as overpayment, underpayment, or duplicate payments. Secondly, it's crucial to verify that the payment and receipt times in the cash flow are later than the task acceptance or result confirmation times in the business flow, ensuring adherence to the business logic of acceptance before payment. Thirdly, it's essential to verify that the payee's identity information matches the task recipient's identity in the business flow, reducing risks such as impersonation and abnormal fund flows.

[0144] When verifying the correlation between cash flow data and invoice flow data, the payment amount, payer, and payee information in the cash flow can be compared one by one with the invoice amount, issuer information, and recipient information in the invoice flow. This ensures that the invoice information matches the cash payment details. Simultaneously, it verifies whether the invoice tax item and tax rate comply with the flexible employment service type and tax regulations. Furthermore, it checks whether the invoice issuance and authentication times conform to the contractual agreement and chronological logic. For example, if it is agreed that the invoice will be issued within 3 days after payment, the invoice issuance time should be after the payment time, thus preventing tax risks such as issuing invoices before payment, not issuing invoices, or issuing false invoices.

[0145] Build a complete data link from contract signing to invoice issuance, and identify whether there are missing or disconnected data gaps in the entire data link.

[0146] In some embodiments, a complete data link is constructed, encompassing contract signing, task assignment, business execution, acceptance confirmation, fund payment, and invoice issuance. Data correlation is verified at each stage based on personnel and task identifiers to identify issues such as missing data, inability to correlate data, or logical gaps. Examples include situations where there is a contract but no business transaction, a business transaction but no funds, funds but no invoice, or data at different stages cannot be correlated using primary keys. Simultaneously, the timestamps of data at each stage are verified to ensure they conform to the business process sequence, guaranteeing a complete, closed-loop data link with no missing critical nodes.

[0147] In this way, by covering multiple dimensions of verification rules such as amount, time sequence, subject, information and link integrity, a closed-loop verification system for the whole process is formed. It can comprehensively investigate problems such as inconsistencies in the four flows, data gaps, and logical conflicts, identify compliance risks such as false business, data fraud and false invoices, and strengthen the whole-link risk control of flexible employment projects.

[0148] In some specific embodiments, optionally, cross-validation is performed on the multimodal task completion credentials in the business flow data, including at least one of the following: verifying whether the location data during task execution matches the specified location, verifying whether the task result image contains key task identifiers, and identifying suspected forged credentials that are duplicated or unrelated to the task.

[0149] Specifically, for multimodal task completion vouchers in business flow data, the authenticity and validity of the vouchers are confirmed through multi-dimensional cross-verification to eliminate the risk of forgery and tampering. For example, the service location specified when the task was assigned (such as the customer address or project site location) can be retrieved and compared with the location data collected during task execution (such as the GPS location of service personnel's mobile phones and equipment location records) to verify whether the location data falls within a reasonable range of the specified location, such as within a radius of 500 meters. Secondly, the uploaded photos, videos, and other image vouchers after the task is completed are identified to verify whether the image vouchers contain key task identifiers, such as customer address, task work order number, exclusive service identifier, and characteristics of equipment to be serviced or already serviced. The consistency between the image shooting time, environment, and task execution scenario is verified to ensure that the image vouchers correspond to the target task and are not reused vouchers from other tasks. Thirdly, suspected forged vouchers can be identified and verified, such as identifying duplicate vouchers and vouchers unrelated to the task, while detecting whether the vouchers have traces of tampering (such as modification of time watermarks, image splicing, etc.), and marking suspected forged vouchers.

[0150] In this way, multimodal voucher cross-validation covers multiple dimensions of location, image, and voucher compliance, which can accurately identify fake vouchers, duplicate vouchers, and tampered vouchers, thereby reducing the risk of relying solely on fake vouchers to process transactions without actual business operations and improving the accuracy of business authenticity verification.

[0151] According to some embodiments of this application, the risk control requirements may optionally include a fault description of the business system. The business system includes, but is not limited to, the FEP system, the CRM system, and / or the BDS system. For example, the fault description of the business system could be that the flexible employment payroll settlement interface cannot be called normally, indicating a connection timeout, etc.

[0152] Figure 7 This is a schematic diagram of the execution flow of a technical expert intelligent agent provided in an embodiment of this application. Figure 7As shown, according to some embodiments of this application, optionally, the technical expert intelligent agent may be used to perform the following steps S701 to S705.

[0153] S701: Perform intent parsing on the fault description to determine the fault type; wherein the fault type includes at least one of interface fault, server fault, or data transmission fault.

[0154] The technical expert agent can perform intent parsing on the fault description submitted by the user to determine the fault type. The fault type can include at least one of the following: interface fault, server fault, or data transmission fault. For example, taking the fault description of the flexible employment payroll settlement interface failing to be called normally and displaying a connection timeout message as an example, after parsing by the technical expert agent, the fault type is determined to be an interface fault.

[0155] S702: Obtain system operation logs, resource configuration information, and the current status of the business system from the business system.

[0156] The technical expert agent can obtain system operation logs, resource configuration information, and the current status of the business system from the malfunctioning business system. The resource configuration information may include server CPU, memory, and disk configuration information, interface parameter configuration information, data transmission protocol and / or bandwidth configuration information, etc.

[0157] S703: Input the fault type, system operation log, resource configuration information, historical fault cases and the current status of the business system into the fine-tuned fault root cause analysis model to obtain the fault root cause analysis results and repair solutions.

[0158] The root cause analysis model can be fine-tuned from a general model using failure cases from business systems, and it possesses scenario-based failure analysis capabilities. In S703, a technical expert agent can input the failure type, system operation logs, resource configuration information, historical failure cases of similar failures, and the current state of the business system into the fine-tuned root cause analysis model. Based on the failure type, the model comprehensively analyzes and infers from the system operation logs, resource configuration information, historical failure cases of similar failures, and the current state of the business system to locate the root cause of the failure and generate a repair plan for that root cause. The repair plan can include operation steps, the implementing entity, required tools, and expected repair time.

[0159] S704: Conduct cost calculations and risk assessments for the repair plan.

[0160] The technical expert agent can assess the development workload or operational costs required to implement a remediation plan, such as 30 minutes of work time for one operations and maintenance personnel. It can also assess the hardware and / or software resource costs required to implement the remediation plan. Furthermore, the agent can evaluate the potential impact of the remediation plan on business systems during implementation, such as the scope of service interruption, the duration of business downtime, and the risk of data loss, for example, a one-hour downtime affecting payroll processing.

[0161] S705: Integrates the root cause analysis results, repair plans, cost calculation results, and risk assessment results to generate and output fault diagnosis results.

[0162] The technical expert agent can structurally integrate root cause analysis results, remediation plans, cost calculation results, and risk assessment results, generating and outputting fault diagnosis results according to preset templates. In some examples, fault diagnosis results may include basic fault information (such as fault type, associated modules, and scope of impact), root cause and analysis basis, detailed content of each remediation plan, and cost calculation details. Fault diagnosis results can be output in the form of visual reports, supporting maintenance personnel and managers to quickly view and make decisions.

[0163] In this way, the entire fault diagnosis process can be automated through the technical expert intelligent agent, from fault description parsing, data retrieval, root cause analysis, solution generation to cost and risk assessment. This replaces the tedious work of manually checking logs and comparing cases one by one, improving fault handling efficiency and reducing business downtime. In addition, based on the finely tuned fault root cause analysis model, combined with multi-dimensional system data and historical fault cases, the root cause of the fault can be accurately located, reducing the subjective bias and root cause misjudgment of manual diagnosis, improving the success rate of fault repair, and reducing ineffective maintenance operations.

[0164] According to some embodiments of this application, risk control requirements may optionally include statistical requirements input by the user in natural language. For example, statistical requirements may include: calculating the labor costs and revenue share of flexible staffing projects in each region during the first quarter of 2025; summarizing the task completion volume and settlement amount of campus operation and maintenance projects in the past three months by service type; or calculating and ranking the monthly payment fulfillment rate of each partner company, etc. This application does not limit these requirements.

[0165] Figure 8 This is a schematic diagram illustrating the execution flow of a data statistics intelligent agent provided in an embodiment of this application. Figure 8 As shown, the data statistics agent can be used to perform the following steps S801 to S805.

[0166] S801: Analyze statistical requirements and generate structured statistical parameters.

[0167] Data statistical intelligence agents can recognize the intent of users' statistical requests input in natural language, extract the core elements of the statistical requests, and transform these core elements into structured statistical parameters that can be recognized by machines. These statistical parameters can include the statistical time range, statistical objects, statistical indicators, grouping dimensions, sorting conditions, and / or filtering conditions, etc.

[0168] S802: Collect valid data that matches the statistical parameters from the business system.

[0169] Based on statistical parameters, raw business data matching the statistical parameters can be collected from the business system. The collected raw business data undergoes validity verification, such as removing null values, outliers, duplicate values, and redundant data irrelevant to the statistical requirements, to obtain valid data matching the statistical parameters.

[0170] S803: Input statistical parameters into the natural language to SQL generation model to generate data statistical statements. By executing the data statistical statements, aggregate and calculate the valid data according to the grouping dimensions specified by the user to obtain statistical results.

[0171] The data statistics agent can input statistical parameters into a natural language to SQL generation model. This model can then generate standardized statistical statements (SQL statements) adapted to the platform's database. The data statistics agent then executes these statements, performing aggregation calculations on the valid data obtained from S802 according to user-specified grouping dimensions, such as summation, average, count, and percentage. Simultaneously, it refines the calculation results by combining sorting and / or filtering conditions to obtain statistical results tailored to the user's needs.

[0172] S804: Match the data type of the statistical results to the corresponding visualization chart template.

[0173] Based on preset chart template matching rules, corresponding visualization chart templates can be matched to statistical results with different characteristics, ensuring that the charts can intuitively and clearly present the core information of the statistical results. The visualization chart templates can include bar charts, line charts, pie charts, donut charts, tables, heatmaps, and / or radar charts, etc., and this application does not limit the types of templates.

[0174] S805: Fill the statistical results into the visualization chart template to obtain the visualization chart.

[0175] According to the preset field mapping rules, the statistical results obtained by S803 are automatically filled into the visualization chart template to obtain the visualization chart.

[0176] In this way, the data statistics intelligent agent can support users to input statistical needs in natural language, without requiring users to master professional SQL statements, statistical tools, or code operations. This lowers the barrier to entry for using the platform's data statistics, adapts to the statistical needs of non-technical personnel (such as operations and management personnel), and improves the universality of the platform's data services. In addition, the entire process of the data statistics intelligent agent can achieve automated data collection and calculation, and can quickly respond to various temporary and personalized statistical needs. It provides timely and accurate data support for the flexible employment platform's operational decisions, risk control analysis, and project management, while also ensuring the traceability of the statistical process and the reusability of the statistical results.

[0177] According to some embodiments of this application, optionally, S801: parsing the statistical requirements and generating structured statistical parameters may include steps six to eight.

[0178] Step 6: Extract the statistical objects, time range, statistical indicators, and grouping dimensions from the statistical requirements.

[0179] The statistical subjects are defined as follows: ...

[0180] Step 7: Map the extracted statistical indicators to the preset statistical indicator dictionary to determine the calculation rules for each statistical indicator.

[0181] The statistical indicator dictionary contains calculation rules for various commonly used statistical indicators in flexible employment scenarios, such as labor costs, task completion volume, return on investment, and contract fulfillment rate. In step seven, the extracted statistical indicators can be mapped to the preset statistical indicator dictionary to determine the calculation rules for each statistical indicator.

[0182] Step 8: Convert the statistical objects, time range, statistical indicators, grouping dimensions, and calculation rules of each statistical indicator into a pre-formatted statistical parameter structure to obtain structured statistical parameters.

[0183] In step eight, the statistical objects, time ranges, statistical indicators, grouping dimensions, and calculation rules for each statistical indicator can be converted into a statistical parameter structure in a preset format (such as JSON / structured data table format) to facilitate machine recognition and execution, thereby obtaining structured statistical parameters.

[0184] In this way, the process of parsing statistical requirements requires no human intervention and is entirely automated by the data statistics intelligence agent. This not only reduces the subjective bias of manual parsing but also improves the efficiency of statistical parameter generation, ensuring a rapid response from requirement input to parameter generation. Furthermore, by introducing a statistical indicator dictionary and achieving precise mapping between indicators and calculation rules, the calculation methods for various statistical indicators in flexible employment scenarios are unified. This solves the statistical bias problems caused by ambiguous natural language expressions and inconsistent manual statistical methods, ensuring consistent calculation results for the same indicator across different statistical requirements and improving the standardization of statistical data.

[0185] According to some embodiments of this application, the multi-agent collaborative risk control system may optionally include a human-computer interaction module. The human-computer interaction module can interact with multiple agents through a visual interface, enabling efficient collaboration between humans and agents, and supporting the generation, interaction optimization, and final delivery of flexible employment risk control reports.

[0186] In some specific embodiments, the human-computer interaction module can be used to perform steps nine through eleven.

[0187] Step 9: Provide a manual review and interactive interface to display the flexible employment risk control report.

[0188] The human-computer interaction module can provide a visual interface for manual review and interaction, which can fully display the flexible employment risk control report generated by the multi-agent collaborative risk control system.

[0189] Step 10: In response to user inquiries or suggestions for modification regarding the flexible employment risk control report, the semantic matching model is invoked to identify the user's intent, and the corresponding working agent is called to adjust the content of the flexible employment risk control report.

[0190] Figure 9 This is a schematic diagram of an architecture for a multi-agent collaborative risk control system for flexible workforce, provided in an embodiment of this application. Figure 9 As shown, in step ten, when a user enters follow-up questions or modification opinions on the risk control report in the manual review and interaction interface, such as supplementing the cross-regional policy adaptation risk analysis of the project or adjusting the quantitative scoring basis of payment default risk, the semantic matching model can be called to identify the user's intent, and based on the user's intent, the corresponding working intelligent agent can be called to supplement or correct the corresponding content of the flexible employment risk control report.

[0191] Step 11: Update and display the flexible employment risk control report in real time on the interface after multiple rounds of interactive optimization.

[0192] The human-computer interaction module can receive adjustments returned by various working agents in real time and integrate the supplemented or corrected content into the flexible employment risk control report according to a preset format, thus achieving dynamic updates to the report. In some embodiments, the updated flexible employment risk control report can retain the original module structure and risk labels, while adding interaction optimization records to show the user intent, executing agent, and adjustment content for each adjustment. This makes the iteration process of the flexible employment risk control report clear and traceable, facilitating subsequent review and auditing, and improving the credibility of the risk control results.

[0193] In this way, through the collaboration of humans and intelligent agents, the flexible employment risk control report can be accurately supplemented and corrected, solving the problem of information omissions or deviations in the initial analysis of the intelligent agent, making the flexible employment risk control report more in line with the actual business scenario and meeting the personalized needs of users.

[0194] like Figure 9 As shown, according to some embodiments of this application, optionally, after the user confirms that the flexible employment risk control report is correct, the multi-agent collaborative risk control system can automatically trigger the knowledge accumulation process of the flexible employment risk control report, marking the report as verified and reusable valid knowledge. Then, the flexible employment risk control report is standardized with graphics and text and stored in the knowledge base.

[0195] like Figure 9 As shown, according to some embodiments of this application, optionally, the multi-agent collaborative risk control system can scan the content of the knowledge base according to a preset period (such as quarterly / semi-annual), and identify outdated knowledge through time validity verification and business adaptability assessment, such as expired policy documents, historical cases that are no longer suitable for the current business model, and model conclusions that have been verified as incorrect. Then, the multi-agent collaborative risk control system can delete outdated knowledge from the knowledge base to avoid invalid knowledge interfering with the analysis and judgment of the agents.

[0196] According to some embodiments of this application, optionally, the multi-agent collaborative risk control system can quantitatively evaluate the performance of each agent (such as policy expert agents, legal and contract agents, etc.) based on model performance evaluation data to pinpoint deviations and deficiencies in the agents or strategies. The model performance evaluation data may include risk identification accuracy, problem handling success rate, user satisfaction, etc., to pinpoint deviations and deficiencies in the model or strategy. Then, for the problems identified in the performance evaluation, the multi-agent collaborative risk control system can perform targeted optimizations, such as adjusting and optimizing the agent's prompts to enhance their scenario adaptability and accuracy; correcting contextual content processing strategies to optimize information retention and association logic in multi-turn dialogues, improving the quality of agent interaction responses; and / or updating model training samples with newly accumulated knowledge to continuously iterate the analytical capabilities of each agent.

[0197] Based on the multi-agent collaborative risk control system 10 for flexible employment provided in any of the above embodiments, this application also provides a multi-agent collaborative risk control method for flexible employment, which can be implemented based on the multi-agent collaborative risk control system 10 for flexible employment as described in any of the above embodiments.

[0198] like Figure 2 As shown in the embodiments of this application, the multi-agent collaborative risk control method for flexible employment may include the following steps: S201: Receive user input of risk control requirements for flexible employment scenarios. Risk control requirements include at least one of the following: compliance review, contract review, cost calculation, risk screening, and data statistics. S202: Based on the fine-tuned semantic matching model, perform intent recognition on risk control requirements; if the risk control requirements are found to contain multiple intents, then combine the multiple intents and decompose the risk control requirements into multiple sub-tasks. S203: Based on a preset routing strategy, the decomposed sub-tasks are distributed to the corresponding working agents, which then execute the respective sub-tasks. Among them, the policy expert agent is used to assess the policy compliance of the flexible employment project; the legal and contract agent is used to conduct risk review of the contract texts of the flexible employment project; the financial advisor agent is used to assess the financial health of the flexible employment project; the project operation agent is used to verify the consistency of the four-stream data of the flexible employment project; the technical expert agent is used to conduct root cause analysis of the flexible employment platform; and the data statistics agent is used to conduct data statistical analysis of the flexible employment project. S204: Monitor the task execution status of each working agent through the supervisor agent, and integrate the output results of each working agent after all sub-tasks are completed to generate a flexible employment risk control report.

[0199] The specific processes of each of the above steps have been described in detail above and will not be repeated here.

[0200] The multi-agent collaborative risk control method for flexible staffing provided in this application, on the one hand, achieves specialized division of risk control tasks through a collaborative architecture of a supervisory agent and multiple vertical domain working agents, effectively solving the efficiency bottleneck problem caused by traditional cross-departmental serial collaboration. Specialized agents such as policy experts, legal and contract / financial advisors perform their respective duties, while the supervisory agent monitors the entire process, enabling orderly scheduling and result integration of multi-agent collaborative work. This improves the timeliness of risk control response and reduces secondary risks such as litigation and arbitration losses, and tax rating downgrades caused by delayed responses. On the other hand, through a fine-tuned semantic matching model and preset routing strategies, complex risk control needs can be accurately identified and broken down into dedicated sub-tasks, which are then distributed to corresponding specialized agents for processing. This avoids the shortcomings of single manual or general-purpose models in covering multiple domain professional needs. Each working agent focuses on a specific domain and can accurately complete professional tasks such as policy compliance assessment, four-stream data consistency verification, and root cause analysis of failures. This effectively reduces risk omissions caused by differences in personnel professional competence and knowledge gaps, improving the accuracy and stability of risk control decisions.

[0201] It should be noted that the multi-agent collaborative risk control method for flexible employment provided in this application embodiment has the same or corresponding technical features as the multi-agent collaborative risk control system for flexible employment provided in any of the above embodiments, and can produce the same technical effects. For the sake of brevity, further details are omitted here.

[0202] The flowcharts and / or block diagrams of methods, systems, and computer program products according to embodiments of this application have been exemplarily described above, and related aspects have been described. It should be understood that each block or combination thereof in the flowcharts and / or block diagrams may be implemented by computer program instructions, by dedicated hardware performing a specified function or action, or by a combination of dedicated hardware and computer instructions. For example, these computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to form a machine such that these instructions, which execute via such processor, enable the implementation of the function / action specified in each block or combination thereof in the flowcharts and / or block diagrams. Such a processor may be a general-purpose processor, a dedicated processor, a special-purpose application processor, or a field-programmable logic circuit.

[0203] The functional blocks shown in the structural block diagrams of this application can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc.; when implemented in software, they are programs or code segments used to perform the required tasks. Programs or code segments can be stored in memory or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. Code segments can be downloaded via computer networks such as the Internet or intranets.

[0204] It should be noted that this application is not limited to the specific configurations and processes described above or shown in the figures. The above descriptions are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the described systems, devices, modules, or units can be referred to the corresponding processes in the method embodiments, and need not be repeated here. It should be understood that the scope of protection of this application is not limited thereto. Any person skilled in the art can conceive of various equivalent modifications or substitutions within the scope of the technology disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application.

Claims

1. A multi-agent collaborative risk control system for flexible employment, the system comprising a supervisory agent and multiple working agents, the multiple working agents including at least two of a policy expert agent, a legal and contractual agent, a financial advisor agent, a project operation agent, a technical expert agent, and a data statistics agent, the system performing the following processes upon startup: Receive user input of risk control requirements for flexible employment scenarios, wherein the risk control requirements include at least one of compliance review, contract review, cost calculation, risk screening, and data statistics; Based on the fine-tuned semantic matching model, the risk control requirement is identified by intent; if the risk control requirement is identified to contain multiple intents, the risk control requirement is decomposed into multiple sub-tasks by combining the multiple intents. Based on a preset routing strategy, the decomposed subtasks are distributed to corresponding working agents, which then execute their respective subtasks. The policy expert AI is used to assess the policy compliance of flexible employment projects; the legal and contract AI is used to conduct risk review of the contract texts of flexible employment projects; the financial advisor AI is used to assess the financial health of flexible employment projects; the project operation AI is used to verify the consistency of the four-stream data of flexible employment projects; the technical expert AI is used to conduct root cause analysis of failures on the flexible employment platform; and the data statistics AI is used to conduct data statistical analysis of flexible employment projects. The supervisory agent monitors the task execution status of each working agent, and after all sub-tasks are completed, the output results of each working agent are integrated to generate a flexible employment risk control report.

2. The system according to claim 1, characterized in that, The system is also used to perform the following processes: Before distributing the decomposed subtasks to the corresponding working agents, determine whether there are any dependencies between the subtasks. If there are no dependencies, each subtask will be distributed to the corresponding working agents for parallel execution. If dependencies exist, the execution order of each subtask is determined, and the corresponding working agents are started sequentially by the supervisor agent according to the execution order, with the output of the upstream working agent serving as the input of the downstream working agent.

3. The system according to claim 1, characterized in that, The risk control requirements include a scenario description for flexible employment projects, and the policy expert intelligent agent is specifically used for: Key feature words are extracted from the scenario description, and a multi-dimensional scenario profile is constructed based on the extracted key feature words. The multi-dimensional profile includes at least two of the following: service scenario, service method, delivery form, settlement method, and personnel scale. Based on the scenario profile, candidate policies applicable to the flexible employment project are selected from the policy library, and the applicable scenario characteristics, core compliance requirements and tax incentive conditions of each candidate policy are analyzed to generate policy feature tags. Calculate the similarity between the scenario profile and the policy feature tags of each candidate policy, and determine the target matching policy based on the similarity matching results; Based on the trained policy interpretation model, the target matching policy is interpreted, and the compliance of the flexible employment project is determined by combining the scenario profile and the policy interpretation results. If the flexible employment project is deemed compliant, the system will match the tax relief policy catalog, verify the eligibility for tax incentives, and analyze the policy basis, and output the policy compliance assessment results.

4. The system according to claim 3, characterized in that, The policy compliance assessment results include at least one of the following: compliance determination conclusion, target matching policy list, tax incentive eligibility determination result, compliance rectification suggestions, and policy risk warning.

5. The system according to claim 1, characterized in that, The aforementioned legal and contract intelligent agents are specifically used for: When risk control requirements include the intent to review contracts, obtain the contract text for flexible employment projects; The contract text structure parsing model is invoked to split the contract document into at least one of the following modules: contract body module, service content module, payment terms module, liability for breach of contract module, data compliance module, confidentiality obligation module, and contract termination module. The content of the terms in each module is semantically encoded to obtain a set of contract terms. The contract clause set is compared with the standard contract template in the contract resource library to identify clauses that are missing, inconsistent with or conflict with the standard contract template. Match the set of contract terms with legal provisions in a legal knowledge base to identify clauses in the set of contract terms that violate mandatory legal provisions and / or have potential legal loopholes; Based on the compliance review results, generate and output the contract risk audit results.

6. The system according to claim 5, characterized in that, The risk control requirements also include user disputes or inquiries regarding the contract documents, and the legal and contract intelligent agent is also used for: Intent identification is performed on the disputed or consulted matters to extract the core demands; By combining contract terms, legal provisions, and user performance history data, the disputed or consulted matters are analyzed to generate a dispute resolution solution; wherein, the dispute resolution solution includes at least one of the following: a negotiation plan, a legal remedy, and a list of evidence; The dispute resolution method will be integrated into the contract risk review results.

7. The system according to claim 1, characterized in that, The financial advisor AI agent is specifically used for: When the risk control requirements include the intent of financial health assessment, obtain the basic data and cost calculation parameters for the flexible staffing project; The basic data and cost calculation parameters are verified, standardized, and missing parameters are filled in to obtain the project calculation parameters. Based on industry benchmark data obtained from the business system, the project calculation parameters are calibrated to obtain the calibrated project calculation parameters. Based on the calibrated project calculation parameters, at least two of the trained net present value calculation model, market value to revenue ratio valuation model, and cost profit margin prediction model are called in parallel to calculate the preliminary financial indicators of the flexible employment project. Sensitivity analysis was conducted on key parameters affecting project profits, and parameter fluctuations were simulated using the Monte Carlo simulation algorithm to obtain sensitivity analysis results, including profit fluctuation ranges and risk warning points. Based on the results of sensitivity analysis, actual financial data of similar historical projects, and customer performance records, the preliminary calculation results are cross-validated to generate an assessment of the financial health status of the flexible employment project.

8. The system according to claim 7, characterized in that, Based on industry benchmark data obtained from the business system, the project calculation parameters are calibrated to obtain the calibrated project calculation parameters, including: The key operating parameters in the project's calculation parameters are compared with similar industry benchmark data, and based on the comparison results, calibration suggestions are generated for key operating parameters that deviate from similar industry benchmark data; The project measurement parameters are updated based on the calibration recommendations to obtain the calibrated project measurement parameters.

9. The system according to claim 7, characterized in that, The assessment results of the financial health status include the calculation results of multiple financial indicators, cost structure analysis table, and conclusions on the financial feasibility of the project. The multiple financial indicators include at least one of the following: net present value of the project, market value to revenue ratio valuation, return on investment, static or dynamic investment payback period, and projected cost and profit details for 1 to 3 years.

10. The system according to claim 1, characterized in that, The risk control requirements include the identification of flexible employment projects and the time frame for risk screening. The project operation intelligent agent is specifically used for: Based on the identification of flexible employment projects and the time range of risk screening, four-flow data related to the flexible employment project are retrieved from multiple business systems. The four-flow data includes contract flow data, business flow data, fund flow data, and invoice flow data. Using personnel identifiers and task identifiers as association primary keys, data packets corresponding to each personnel are constructed based on the four streams of data. Perform four-stream pass-through correlation verification on each data packet; The authenticity of the business is verified in multiple dimensions, including cross-validation of multimodal task completion credentials in the business flow data, verification of the consistency of personnel identity information in each stage, and verification of the rationality of business data based on historical business baseline data. Based on the results of the correlation verification and business authenticity verification, the specific subject and link of the abnormal data are located, and the risk level and impact scope of the abnormality are assessed according to the preset rules. Generate a consistency verification report for the four flows, including a list of abnormal data, risk level classification, and handling guidelines.

11. The system according to claim 10, characterized in that, Perform four-stream pass-through correlation verification on each data packet, including at least one of the following: Perform correlation verification between contract flow data and business flow data to verify whether the task recipient has a valid contract and whether the task scope is consistent, and verify whether the task dispatch time is within the contract validity period. Perform correlation verification on business flow data and fund flow data, verify whether the settlement amount matches the task results, verify whether the payment time is later than the task acceptance time, and verify whether the identity of the payee is consistent. Perform correlation verification between fund flow data and invoice flow data, verify whether the invoice information is consistent with the payment amount and the payment entity, and verify whether the invoice issuance sequence conforms to the agreement; Build a complete data link from contract signing to invoice issuance, and identify whether there are missing or disconnected data gaps in the entire data link.

12. The system according to claim 10, characterized in that, Cross-validate the multimodal task completion credentials in the business flow data, including at least one of the following: verify whether the location data during task execution matches the specified location, verify whether the task result image contains key task identifiers, and identify suspected forged credentials that are duplicated or irrelevant to the task.

13. The system according to claim 1, characterized in that, The risk control requirements include fault descriptions of the business system, and the technical expert agent is specifically used for: The fault description is parsed to determine the fault type; wherein the fault type includes at least one of interface fault, server fault, or data transmission fault. Obtain system operation logs, resource configuration information, and the current status of the business system from the business system; Input the fault type, system operation log, resource configuration information, historical fault cases, and the current status of the business system into the fine-tuned fault root cause analysis model to obtain the fault root cause analysis results and repair solutions. Cost calculations and risk assessments were conducted for the repair plan; The system integrates the root cause analysis results, repair plans, cost calculation results, and risk assessment results to generate and output fault diagnosis results.

14. The system according to claim 1, characterized in that, The risk control requirements include statistical requirements input by the user in natural language, and the data statistics agent is specifically used for: Analyze statistical requirements and generate structured statistical parameters; Collect valid data that matches the statistical parameters from the business system; Statistical parameters are input into the natural language to SQL generation model to generate data statistical statements. By executing the data statistical statements, the effective data is aggregated and calculated according to the grouping dimensions specified by the user to obtain statistical results. Based on the data type of the statistical results, match the corresponding visualization chart template; Fill the statistical results into the visualization chart template to obtain the visualization chart.

15. The system according to claim 14, characterized in that, The process of parsing statistical requirements and generating structured statistical parameters includes: Extract the statistical objects, time ranges, statistical indicators, and grouping dimensions from the statistical requirements; The extracted statistical indicators are mapped to a pre-defined dictionary of statistical indicators to determine the calculation rules for each statistical indicator. The statistical objects, time ranges, statistical indicators, grouping dimensions, and calculation rules of each statistical indicator are converted into a pre-formatted statistical parameter structure to obtain structured statistical parameters.

16. The system according to claim 1, characterized in that, The system also includes a human-computer interaction module for: Provides manual review and interactive interface, and displays flexible employment risk control reports; In response to user input regarding follow-up questions or suggestions for modification to the flexible employment risk control report, the semantic matching model is invoked to identify the user's intent, and the corresponding working agent is invoked to adjust the content of the flexible employment risk control report. The interface updates and displays flexible employment risk control reports in real time, after multiple rounds of interactive optimization.

17. A multi-agent collaborative risk control method for flexible employment, the method being implemented based on the multi-agent collaborative risk control system for flexible employment as described in any one of claims 1 to 16, the method comprising: Receive user input of risk control requirements for flexible employment scenarios, wherein the risk control requirements include at least one of compliance review, contract review, cost calculation, risk screening, and data statistics; Based on the fine-tuned semantic matching model, the risk control requirement is identified by intent; if the risk control requirement is identified to contain multiple intents, the risk control requirement is decomposed into multiple sub-tasks by combining the multiple intents. Based on a preset routing strategy, the decomposed sub-tasks are distributed to corresponding working agents, which then execute their respective sub-tasks. Among them, the policy expert agent is used to assess the policy compliance of the flexible employment project; the legal and contract agent is used to conduct risk review of the contract texts of the flexible employment project; the financial advisor agent is used to assess the financial health of the flexible employment project; the project operation agent is used to verify the consistency of the four-stream data of the flexible employment project; the technical expert agent is used to conduct root cause analysis of the flexible employment platform; and the data statistics agent is used to conduct data statistical analysis of the flexible employment project. The supervisory agent monitors the task execution status of each working agent, and after all sub-tasks are completed, the output results of each working agent are integrated to generate a flexible employment risk control report.

Images