System development-oriented multi-agent collaborative arrangement method, system, device and medium

Abstract

The application discloses a kind of multi-agent collaborative arrangement methods, system, equipment and medium for system development, it is related to artificial intelligence technical field, including: in response to received user request, the intention of user request is identified, to determine target task and the intelligent agent of processing target task, based on target task, reference information is retrieved from knowledge base and is injected into intelligent agent, task processing process information in shared data carrier is read as input by intelligent agent, processing result is generated based on reference information, processing result is written in the data area corresponding to system processing stage in shared data carrier, processing result is sent to user by interactive interface, in response to received user reply information, user reply information is parsed, when user reply information is confirmation, update system processing stage;It has achieved the effect that cross-stage information is automatically inherited, the accuracy and compliance of generated result are improved, the dependence of manual scheduling is reduced, and the user confirmation state can be tracked.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a multi-agent collaborative orchestration method, system, device and medium for system development. Background Technology

[0002] With the rapid development of fintech, the complexity of developing overseas securities trading systems continues to increase. On the one hand, compliance requirements from financial regulatory agencies are becoming increasingly stringent, making compliance risk management during the development process a key challenge. On the other hand, the expansion of R&D teams and frequent cross-team collaborations have made the effective transmission of requirement information and the reuse of historical knowledge bottlenecks that restrict R&D efficiency.

[0003] In recent years, multi-agent systems have achieved the decomposition and parallel processing of complex tasks through the collaboration of multiple specialized intelligent agents. However, the systematic integration and application of these technologies in overseas securities software development scenarios currently lack mature solutions, especially in key areas such as cross-stage information inheritance, proactive injection of compliance knowledge, human-machine collaborative state tracking, and multi-signal risk aggregation, where existing technologies have significant shortcomings. Summary of the Invention

[0004] The main purpose of this application is to provide a multi-agent collaborative orchestration method, system, device and medium for system development, which aims to achieve the effects of automatic cross-stage information inheritance, improved accuracy and compliance of generated results, reduced reliance on manual scheduling and traceable user confirmation status.

[0005] To achieve the above objectives, this application proposes a multi-agent collaborative orchestration method for system development, including: In response to a received user request, intent recognition is performed on the user request to determine the target task and at least one of a plurality of intelligent agents that handle the target task; Based on the key information of the target task, reference information is retrieved from at least one knowledge base and injected into the intelligent agent; The intelligent agent reads the task processing information in the shared data carrier as input, generates a processing result based on the reference information, and writes the processing result into the data area in the shared data carrier corresponding to the system processing stage; wherein, the shared data carrier includes multiple data areas divided according to the system processing stage, and the shared data carrier is used to transmit the task processing information among the multiple intelligent agents; The system sends the processing result to the user through an interactive interface, and in response to the received user reply, it parses the user reply. When the user reply is confirmed, the system updates the processing stage.

[0006] In one embodiment, the step of performing intent recognition on the user request to determine a target task and at least one of a plurality of intelligent agents processing the target task includes: Based on the semantics of the user request, identify the target task corresponding to the user request; Based on the task type of the target task, at least one agent is determined from the plurality of agents to handle the target task.

[0007] In one embodiment, retrieving reference information from at least one knowledge base based on key information of the target task includes: Extract key information of the target task, convert the key information into a query vector, perform semantic retrieval on the at least one knowledge base, and obtain multiple candidate knowledge fragments; The candidate knowledge fragments are sorted and selected based on their relevance. The selected knowledge fragments are combined with the key information of the target task to obtain the reference information.

[0008] In one embodiment, the step of reading task processing information from a shared data carrier as input by an intelligent agent and generating a processing result based on the reference information includes: The intelligent agent determines the data region in the shared data carrier corresponding to the current system processing stage based on the current system processing stage, and obtains the task processing information. The processing result is generated by inputting the task processing information and the injected reference information into the intelligent agent.

[0009] In one embodiment, after responding to the received user reply information, the method further includes: Perform semantic parsing on the user response information to identify the type of the user response information; When the response information is modified, the data area in the shared data carrier corresponding to the current processing stage is updated based on the modification opinion, and the agent is triggered to regenerate the processing result based on the modification opinion; When the response is a rejection, the agent is triggered to re-execute the processing flow corresponding to the target task.

[0010] In one embodiment, the method further includes: Heterogeneous signals collected from multiple heterogeneous signal sources are normalized, and a comprehensive risk score is obtained based on the information already written in the shared data carrier and the normalized heterogeneous signals. The comprehensive risk score triggers a tiered early warning system.

[0011] In one embodiment, after the system processing phase is updated, the process further includes: After the processing results generated by the intelligent agent are structured, they are asynchronously written into the at least one knowledge base; The retrieval weights of information in at least one knowledge base are adjusted based on the adoption of the reference information.

[0012] Furthermore, to achieve the above objectives, this application also proposes a multi-agent collaborative orchestration system for system development, comprising: An intent recognition module is used to, in response to a received user request, recognize the intent of the user request to determine the target task and at least one of a plurality of intelligent agents that process the target task. A knowledge retrieval module is used to retrieve reference information from at least one knowledge base based on key information of the target task, and inject the reference information into the intelligent agent; The intelligent agent execution module is used to read task processing information from the shared data carrier as input through the intelligent agent, generate processing results based on the reference information, and write the processing results into the data area in the shared data carrier corresponding to the system processing stage; wherein, the shared data carrier includes multiple data areas divided according to the system processing stage, and the shared data carrier is used to transmit the task processing information among the multiple intelligent agents; The response processing module is used to send the processing result to the user through an interactive interface, parse the user response information in response to the received user response information, and update the system processing stage when the user response information is confirmation.

[0013] Furthermore, to achieve the above objectives, this application also proposes a multi-Agent collaborative orchestration device for system development, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the multi-Agent collaborative orchestration method for system development as described above.

[0014] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the multi-Agent collaborative orchestration method for system development as described above.

[0015] The present application proposes one or more technical solutions, which have at least the following technical effects: By responding to a received user request, the intent of the user request for reference information is identified to determine the target task and the intelligent agent handling the target task. Based on the key information of the target task, reference information is retrieved from the knowledge base and injected into the intelligent agent. The intelligent agent generates a processing result based on the task processing information and reference information in the shared data carrier. The processing result is written into the data area corresponding to the system processing stage in the shared data carrier. The processing result is sent to the user through an interactive interface. In response to the received user reply, the user reply is parsed. When the user reply is confirmation, the system processing stage is updated. This application eliminates the manual information transmission link by transmitting task processing information among multiple intelligent agents through a shared data carrier. It reduces the compliance risk of the generated result by actively injecting reference information and achieves automatic state transition for human-machine collaborative confirmation by responding to user replies. It transforms user confirmation into a traceable system state, achieving the effects of automatic cross-stage information inheritance, improved accuracy and compliance of generated results, reduced reliance on manual scheduling, and traceable user confirmation status. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the first embodiment of the multi-Agent collaborative orchestration method for system development in this application; Figure 2 This is an optional flowchart of step S101 in the first embodiment of the multi-Agent collaborative orchestration method for system development in this application; Figure 3 This is an optional flowchart of step S102, retrieving reference information, in the first embodiment of the multi-Agent collaborative orchestration method for system development in this application. Figure 4 This is an optional flowchart for step S103 of generating the processing result in the first embodiment of the multi-Agent collaborative orchestration method for system development in this application; Figure 5 This is an alternative flowchart following the received user reply information in the first embodiment of the multi-Agent collaborative orchestration method for system development in this application; Figure 6 This is an optional flowchart for risk warning in the first embodiment of the multi-Agent collaborative orchestration method for system development in this application; Figure 7 This is an optional flowchart following step S104 in the first embodiment of the multi-Agent collaborative orchestration method for system development in this application; Figure 8 This is a schematic diagram of a multi-agent collaborative orchestration system for system development according to an embodiment of this application; Figure 9 This is a schematic diagram of the structure of a multi-agent collaborative orchestration system for system development according to an embodiment of this application; Figure 10 This is a schematic diagram of the structure of a multi-agent collaborative orchestration device for system development, as described in an embodiment of this application. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0020] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0021] The main solution of this application embodiment is as follows: In response to a received user request, the intent of the user request for reference information is identified to determine the target task and the intelligent agent handling the target task. Based on the key information of the target task, reference information is retrieved from the knowledge base and injected into the intelligent agent. The intelligent agent generates a processing result based on the task processing information and reference information in the shared data carrier, writes the processing result into the data area corresponding to the system processing stage in the shared data carrier, and sends the processing result to the user through an interactive interface. In response to the received user reply, the user reply is parsed. When the user reply is confirmation, the system processing stage is updated. This application eliminates the manual information transmission link by transmitting task processing information among multiple intelligent agents through a shared data carrier, reduces the compliance risk of the generated result by actively injecting reference information, and realizes automatic state transition of human-machine collaborative confirmation by responding to user replies. It transforms user confirmation into a traceable system state, achieving the effects of automatic cross-stage information inheritance, improved accuracy and compliance of generated results, reduced reliance on manual scheduling, and traceable user confirmation status.

[0022] With the rapid development of fintech, the complexity of developing overseas securities trading systems continues to increase. On the one hand, compliance requirements from financial regulatory agencies are becoming increasingly stringent, making compliance risk management during the development process a key challenge. On the other hand, the expansion of R&D teams and frequent cross-team collaborations have made the effective transmission of requirement information and the reuse of historical knowledge bottlenecks that restrict R&D efficiency.

[0023] In recent years, large language models, represented by the GPT and DeepSeek series, have made breakthroughs in natural language understanding and generation, providing a technological foundation for the intelligent processing of software requirements. RAG technology, by combining external knowledge bases with large language models, effectively solves the problems of insufficient model knowledge timeliness and lack of domain knowledge. Multi-Agent Systems (MAS) achieve the decomposition and parallel processing of complex tasks through the collaboration of multiple specialized agents. However, the systematic integration and application of these technologies in overseas securities software development scenarios currently lack mature solutions, especially in key areas such as cross-stage information inheritance, proactive injection of compliance knowledge, human-machine collaborative state tracking, and multi-signal risk aggregation, where existing technologies all have significant shortcomings.

[0024] This application provides a solution that transmits task processing information among multiple intelligent agents through a shared data carrier, eliminating the manual information transmission link, reducing compliance risks of generated results by actively injecting reference information, and realizing automatic state transition of human-machine collaborative confirmation by responding to user replies, transforming user confirmation into a traceable system state. This achieves the effects of automatic cross-stage information inheritance, improved accuracy and compliance of generated results, reduced reliance on manual scheduling, and traceable user confirmation status.

[0025] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions, a multi-agent collaborative orchestration system or device developed for system applications, etc. The following description uses a multi-agent collaborative orchestration system developed for system applications as an example to illustrate this embodiment and the subsequent embodiments.

[0026] Based on this, embodiments of this application provide a multi-agent collaborative orchestration method for system development, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the multi-Agent collaborative orchestration method for system development in this application.

[0027] In this embodiment, the multi-agent collaborative orchestration method for system development includes steps S101 to S104. The following provides a detailed explanation of each step.

[0028] The multi-agent collaborative orchestration method for system development includes: Step S101: In response to the received user request, perform intent recognition on the user request to determine the target task and at least one of the multiple intelligent agents that will handle the target task.

[0029] In some embodiments of this application, upon receiving a user request, the intent recognition agent employs a dual-mode recognition mechanism combining large language model semantic understanding and a rule engine to semantically parse the natural language content of the user request and understand the user's intent. For example, it identifies "Help me break down this requirement" as a requirement processing intent, "Test request submitted" as a test processing intent, and "Recent tasks are severely delayed" as a risk warning intent. On the other hand, it performs rapid matching and fallback based on prior rules such as event type, keywords, and message source to ensure the stability and interpretability of the recognition results.

[0030] It should be noted that this embodiment includes, but is not limited to, four types of business intelligent agents: demand intelligent agent, collaborative scheduling intelligent agent, testing intelligent agent, and project early warning intelligent agent.

[0031] In some embodiments of this application, when a target task directly matches the responsibilities of a certain type of intelligent agent, it is directly routed to that intelligent agent. For example, "requirement decomposition" is routed to the requirements intelligent agent, and "test case generation" is routed to the test intelligent agent. When the target task involves cross-stages or complex processes, the intent identification intelligent agent may delegate the task to the collaborative scheduling intelligent agent, which then schedules multiple intelligent agents to execute collaboratively. For example, a complete R&D task may simultaneously require the requirements intelligent agent to complete decomposition, the collaborative scheduling intelligent agent to complete assignment, and the test intelligent agent to prepare test cases. In this case, the collaborative scheduling intelligent agent acts as the "scheduling hub" for unified orchestration.

[0032] It should be noted that by combining semantic understanding and rule engine with contextual information in shared data carrier, the task is ensured to be dispatched to the most suitable agent, and the intention understanding and agent selection are completed automatically, reducing the cost of manual scheduling. When the task is complex, it can be routed to a collaborative scheduling agent for multi-agent orchestration.

[0033] Step S102: Based on the key information of the target task, retrieve reference information from at least one knowledge base and inject the reference information into the agent.

[0034] It should be noted that a multi-database knowledge system has been pre-built for overseas securities research and development scenarios, including: The compliance rules library includes the Securities and Futures Ordinance, the SFC Code of Conduct, the SEC Regulation NMS, the CCASS Operating Procedures, and cross-border settlement rules, among others. The requirements clarification knowledge base includes historical requirements breakdown plans, clarification records, and adopted opinions; The test point recall knowledge base includes historical test points and test strategies; The historical test case knowledge base includes historical test cases and defect patterns; The graph database includes team responsibility graphs and risk graphs.

[0035] In some embodiments of this application, the intelligent agent selects one or more relevant knowledge bases for retrieval based on the target task type. For example, the requirements intelligent agent retrieves a requirements clarification knowledge base and a compliance rule base; the testing intelligent agent retrieves a test point recall knowledge base, a historical use case knowledge base, and a compliance rule base. During the retrieval process, key information of the target task is converted into a vector representation. Semantic similarity retrieval is performed in parallel on multiple relevant knowledge bases using the query vector to recall knowledge fragments semantically similar to the query vector. The recalled candidate knowledge fragments are ranked by relevance, and multiple most relevant document fragments are selected as reference information. The retrieved reference information is combined with the original input of the target task and concatenated into the system prompt words or inference context of the large language model used by the intelligent agent. For example, for the requirements intelligent agent, the reference information is organized together with the original requirements description into prompt words, enabling the large language model to refer to historical experience and compliance requirements when generating task cards.

[0036] It should be noted that retrieving reference information from at least one knowledge base eliminates the need for manual searching or providing of domain knowledge, thus automating knowledge acquisition. Step S103: The intelligent agent reads the task processing information in the shared data carrier as input, generates the processing result based on the reference information, and writes the processing result into the data area in the shared data carrier corresponding to the system processing stage.

[0037] It should be noted that task processing information refers to historical data related to the current target task that has been stored in the shared data carrier and comes from the previous processing stage. It is written by the previous intelligent agent during the processing and is the direct basis for the current intelligent agent to understand the task background, inherit the previous results, and avoid duplication of work.

[0038] In some embodiments of this application, the agent determines the current processing stage based on routing information and task type. It then searches the shared data carrier for the data region corresponding to the current processing stage. From this data region and possible preceding data regions, it extracts information related to the current target task, such as preceding processing results, confirmation records, and modification suggestions. This information, along with injected reference information and the original target task information, serves as the input basis for the agent's reasoning. For example, when generating test cases, the test agent reads the task card set and acceptance criteria from the requirements stage data region of the shared data carrier, as well as the confirmation records from the collaboration stage data region. This eliminates the need to re-understand the requirements, allowing for direct test design based on preceding results. Based on these two types of information and the target task itself, a structured processing result is generated through its reasoning model. The specific content and format of the processing result depend on the target task type. After generating the processing result, the agent immediately writes it into the shared data carrier. During writing, the agent writes the processing result into the corresponding data region according to the current system processing stage and may simultaneously update relevant state fields.

[0039] It should be noted that by reading the preceding processing information in the shared data carrier, the subsequent intelligent agent seamlessly inherits the results of the preceding process, eliminating the manual information transmission link. At the same time, by utilizing reference information and preceding context, the generated results are more in line with the actual task background and domain requirements. The processing results are written in real time to form a complete historical record, supporting version backtracking and status tracing.

[0040] Step S104: Send the processing result to the user through the interactive interface, respond to the received user reply information, parse the user reply information, and update the system processing stage when the user reply information is confirmation.

[0041] It should be noted that after the intelligent agent generates the processing result, it pushes the processing result to the user through the interaction interface. After the user receives the confirmation message, he / she replies through the interaction interface. After receiving the reply, the agent performs structured semantic parsing on the reply content to identify the user's intent.

[0042] It should be noted that user response types may include, but are not limited to: confirmation (acknowledging the processing result); modification (suggesting modifications); and rejection (denying the processing result).

[0043] It should be noted that upon receiving confirmation from the user, the current task will be moved from the current processing stage (e.g., requirement breakdown pending confirmation) to the next processing stage (e.g., requirement being converted to development) to trigger the initiation of subsequent processes.

[0044] Furthermore, if the user does not respond with a confirmation message within a preset time, a multi-level escalation notification will be automatically triggered: first, the responsible person will be notified; second, the project manager will be notified; and third, the department head will be notified. This mechanism ensures that key confirmation nodes will not be indefinitely stalled due to human delays, guaranteeing the continuous progress of the R&D process.

[0045] It should be noted that parsing the user's "confirmation" behavior as a structured event automatically triggers updates in the system's processing phase, making the results of manual confirmation traceable and auditable, eliminating the problem of verbal confirmation being unrecordable and untraceable; the processing results generated by the intelligent agent must be confirmed by the user before proceeding to the next stage, ensuring that AI automated decision-making is constrained by human review at key stages, preventing unreviewed results from directly entering the next stage.

[0046] Reference Figure 2 As shown, Figure 2 This is an optional flowchart of step S101 in the first embodiment of the multi-Agent collaborative orchestration method for system development in this application. The method may include, but is not limited to, steps S201 to S202.

[0047] Step S201: Based on the semantics of the user request, identify the target task corresponding to the user request.

[0048] In some embodiments of this application, upon receiving a user request, the intent recognition agent employs a dual-mode recognition mechanism combining large language model semantic understanding and a rule engine to semantically parse the natural language content of the user request and understand the user's intent. For example, it identifies "Help me break down this requirement" as a requirement processing intent, "Test request submitted" as a test processing intent, and "Recent tasks are severely delayed" as a risk warning intent. On the other hand, it performs rapid matching and fallback based on prior rules such as event type, keywords, and message source to ensure the stability and interpretability of the recognition results.

[0049] It should be noted that, based on understanding the semantics of user requests, the system maps the semantics to standardized target task types. The identified target task provides a clear operational object for subsequent steps: the intent recognition agent knows "what to do" in order to further determine "who should do it".

[0050] Step S202: Based on the task type of the target task, determine at least one agent from multiple agents to handle the target task.

[0051] It's important to note that task type is the most direct basis for determining the agent. For example, when the target task is "requirement decomposition," the requirement agent is directly matched; when the target task is "test case generation," the testing agent is directly matched. On the other hand, the same task type may be handled by different agents at different processing stages. For instance, in the initial requirement phase, the requirement agent handles requirement decomposition, while task assignment after requirement confirmation is handled by the coordination and scheduling agent. Decision-making by combining multiple factors such as task type, current stage, and historical information ensures that tasks are assigned to the most suitable execution agent, avoiding rework caused by incorrect routing.

[0052] Reference Figure 3 As shown, Figure 3 This is an optional flowchart of step S102, which is used to retrieve reference information, in the first embodiment of the multi-Agent collaborative orchestration method for system development in this application. The method may include, but is not limited to, steps S301 to S303.

[0053] Step S301: Extract key information of the target task, convert the key information into query vectors, perform semantic retrieval on at least one knowledge base, and obtain multiple candidate knowledge fragments.

[0054] It should be noted that in the process of extracting key information from the target task, the above elements can be extracted from the original task description based on rule templates, keyword extraction algorithms, or large language model understanding, forming structured query fields. For example, for a requirement decomposition task, key information such as "dark pool trading" and "risk control verification" may be extracted; for a test case generation task, "acceptance criteria" and "functional points" in the task card may be extracted as key information.

[0055] It should be noted that converting key information into query vectors involves using an embedding model to map key information in text form into numerical vectors in a high-dimensional space. This makes texts with similar semantics closer together in the vector space, while texts with different semantics are farther apart.

[0056] In some embodiments of this application, multiple extracted key information fields are concatenated into a query text in a certain order, and normalization processing such as capitalization and special character correction is performed. A pre-trained embedding model is used to encode the query text, outputting a fixed-dimensional vector. One or more relevant knowledge bases are selected based on the target task type. For example, a requirement agent can select a requirement clarification knowledge base and a compliance rule base; a testing agent can select a test point recall knowledge base, a historical use case knowledge base, and a compliance rule base. Having at least one knowledge base means that multiple bases can be retrieved in parallel to cover different types of reference information. The knowledge entry whose semantics are closest to the query vector is searched in the vector index of the knowledge base.

[0057] It should be noted that candidate knowledge fragments are the preliminary recall results of semantic retrieval, which are usually several records or document fragments in the knowledge base that rank highly similar to the query vector.

[0058] Step S302: Sort and select multiple candidate knowledge fragments based on their relevance.

[0059] It should be noted that the candidate knowledge fragments returned by semantic retrieval all have a certain degree of semantic similarity to the query vector, but the degree of similarity is not equal. Some fragments may only be partially relevant or belong to the adjacent domain but are not directly applicable. Therefore, it is necessary to rank the candidate fragments according to the relevance index, so that the fragments that best match the current target task are ranked first. By ranking and selecting, weakly relevant and low-quality fragments are eliminated, ensuring that the reference information injected into the agent is highly relevant to the target task and reducing noise interference.

[0060] Step S303: Combine the selected knowledge fragments with the key information of the target task to obtain reference information.

[0061] It should be noted that the selected knowledge fragments are organized and integrated with the key information of the target task itself to form a complete, semantically coherent reference information that can be directly used as the basis for the agent's reasoning. This allows the agent to obtain the support of domain knowledge while maintaining accurate focus on the target task itself during reasoning.

[0062] It should be noted that the combined reference information includes two parts: a clear description of the current task and relevant reference materials from the knowledge base. The two complement each other and form an organic whole.

[0063] Reference Figure 4 As shown, Figure 4 This is an optional flowchart for step S103 of the first embodiment of the multi-Agent collaborative orchestration method for system development in this application, which generates the processing result. The method may include, but is not limited to, steps S401 to S402.

[0064] Step S401: The agent determines the data region in the shared data carrier that corresponds to the current system processing stage based on the current system processing stage, and obtains the task processing process information.

[0065] It should be noted that the system processing stage refers to the stage division in the multi-agent collaborative processing flow, such as the requirement stage, collaboration stage, testing stage, and early warning stage. The current system processing stage is determined by the agent based on routing information and task type, and usually corresponds to the development stage to which the target task belongs.

[0066] It should be noted that the shared data carrier is structurally divided into multiple data regions according to the system processing stages. Each data region stores task processing information related to that stage. The data region corresponding to the current system processing stage refers to the region in the above division that matches the stage the agent is currently in. For example, if the current processing stage of the test agent is the testing stage, then the corresponding data region is the testing stage data region.

[0067] In some embodiments of this application, the agent determines the identifier of the current system processing stage based on routing information and task type. In the structure of the shared data carrier, the agent locates the corresponding data area based on the stage identifier. Since the shared data carrier provides a unified read / write interface, the agent can initiate a read request through this interface with the stage identifier as a parameter. The management component of the shared data carrier returns the content of the corresponding data area without the agent having to traverse all the data itself.

[0068] Step S402: Generate processing results by inputting task processing information and injected reference information into the agent.

[0069] It should be noted that by using the above two types of information as the inference context input of the large language model used by the intelligent agent, the large language model can see both the preceding context and domain knowledge when generating the processing results, rather than just inferring based on the user's original request.

[0070] In some embodiments of this application, task processing information and reference information are combined with the original description of the target task and organized into a complete reasoning context according to a preset prompt word template. The agent inputs the above reasoning context into a large language model. The large language model performs reasoning based on its pre-trained knowledge and the information in the context, generates a preliminary output of the processed content, and performs structured processing on the preliminary output to obtain a processing result that conforms to a preset format.

[0071] Reference Figure 5 As shown, Figure 5 This is an alternative flowchart following the received user reply information in the first embodiment of the multi-Agent collaborative orchestration method for system development in this application. The method may include, but is not limited to, steps S501 to S503.

[0072] Step S501: Perform semantic parsing on the user's reply information to identify the type of user's reply information.

[0073] It should be noted that after the intelligent agent generates the processing result, it pushes the processing result to the user through the interaction interface. After the user receives the confirmation message, he / she replies through the interaction interface. After receiving the reply, the agent performs structured semantic parsing on the reply content to identify the user's intent.

[0074] It should be noted that user response types may include, but are not limited to: confirmation (acknowledging the processing result); modification (suggesting modifications); and rejection (denying the processing result).

[0075] Step S502: When the reply is "modification", update the data area in the shared data carrier corresponding to the current processing stage based on the modification opinion, and trigger the agent to regenerate the processing result based on the modification opinion.

[0076] It should be noted that the system writes the extracted modification suggestions into the data area corresponding to the current processing stage in the shared data carrier, making it recordable and traceable official data. When the data area corresponding to the current processing stage in the shared data carrier is updated, the modification suggestions are added as new information or associated with the original processing result, rather than being deleted or replaced. The original processing result is still retained in the shared data carrier, forming a historical version record.

[0077] In some embodiments of this application, the modification opinions are read by the intelligent agent as newly added task processing information, so that the intelligent agent clearly knows which content the user is dissatisfied with and how he / she expects to adjust it. The regenerated processing results are also written into the data area corresponding to the current processing stage in the shared data carrier to form a new version. The new version is associated with the old version through the version number, which supports comparison and backtracking.

[0078] Step S503: When the reply message is rejection, the agent is triggered to re-execute the processing flow corresponding to the target task.

[0079] It should be noted that the processing flow corresponding to the target task refers to the complete process from the start of the target task to the generation of the processing result, not just the regeneration of the result itself. The original processing result will not be physically deleted, but will be retained in the shared data carrier, but will be marked as "rejected" or "invalid", and will be associated with the newly generated version through a version number.

[0080] Reference Figure 6 As shown, Figure 6 This is an optional flowchart of risk warning in the first embodiment of the multi-Agent collaborative orchestration method for system development in this application. The method may include, but is not limited to, steps S601 to S602.

[0081] Step S601: Collect heterogeneous signals from multiple heterogeneous signal sources, normalize the heterogeneous signals, and obtain a comprehensive risk score based on the information already written in the shared data carrier and the normalized heterogeneous signals.

[0082] It should be noted that heterogeneous signals refer to risk-related signals from different systems with different data formats and semantics. The process of mapping heterogeneous signals to a common numerical range makes signals from different sources and with different dimensions comparable and aggregateable.

[0083] It should be noted that heterogeneous signals provide a complete project context, allowing the scoring to be based not only on real-time signals but also on the progress of previous phases and historical risk patterns. For example, if the shared data carrier shows that the requirement phase has experienced multiple rejections, it indicates an unstable understanding of the requirements, which would amplify the risk weight of the current delay signal or include it as an additional risk factor in the scoring.

[0084] Step S602: Trigger a tiered early warning based on the comprehensive risk score.

[0085] It should be noted that tiered early warning refers to dividing early warnings into multiple levels, with different levels corresponding to different degrees of risk severity and response strategies.

[0086] In some embodiments of this application, the warning levels include, but are not limited to, a three-level classification: A yellow alert will be issued when progress is slightly delayed (delay rate <10%) or quality fluctuates slightly (defect rate <5%), and a notification will be pushed to the responsible person through the interactive interface. Orange alert: When progress is significantly delayed (delay rate 10%-30%) or compliance risks emerge (there are unmet compliance check items), a notification and email will be sent, along with suggested handling solutions. A red alert is issued when there is a serious delay (delay rate > 30%) or a major compliance violation (violation of key regulatory provisions). This triggers an emergency escalation process, including notification and phone calls, and notifies the department head.

[0087] Reference Figure 7 As shown, Figure 7 This is an optional flowchart following step S104 in the first embodiment of the multi-Agent collaborative orchestration method for system development in this application. The method may include, but is not limited to, steps S701 to S702.

[0088] Step S701: After structuring the processing results generated by the agent, asynchronously write them into at least one knowledge base.

[0089] It should be noted that the processing results generated by the intelligent agents vary in format and content depending on the task type. For example, a requirements intelligent agent generates a set of task cards and a task dependency graph; a testing intelligent agent generates a set of structured test cases; and a project early warning intelligent agent generates risk scores, warning levels, and handling suggestions. Although these results are structured to some extent, their field definitions, granularity, and relationships are not uniform, making them unsuitable for direct input into the knowledge base for semantic retrieval.

[0090] It's important to note that asynchronous writing means the knowledge backflow operation does not block the main execution of the agent. After generating the processing result, the agent can continue with subsequent operations, such as sending confirmation requests to the user or advancing the stage status, without waiting for the knowledge base to be written completely. The write operation is executed independently in the background by the system, triggered by a message queue or asynchronous task mechanism.

[0091] Step S702: Adjust the retrieval weight of information in at least one knowledge base based on the adoption of reference information.

[0092] It's important to note that retrieval weight is a key parameter influencing the ranking of semantic retrieval results. Semantic retrieval ranking not only depends on the similarity score between the query vector and the knowledge item vector, but also incorporates weighting factors to adjust the ranking. Higher weights result in a higher ranking for the knowledge item in the retrieval results, making it more likely to be selected and injected into the agent.

[0093] This application also provides a multi-agent collaborative orchestration system for system development; please refer to [reference needed]. Figure 8 The multi-agent collaborative orchestration system for system development includes: The intent recognition module 801 is used to recognize the intent of a received user request in response to the user request, so as to determine the target task and at least one of a plurality of intelligent agents for processing the target task. The knowledge retrieval module 802 is used to retrieve reference information from at least one knowledge base based on key information of the target task, and inject the reference information into the intelligent agent; The agent execution module 803 is used to read the task processing information in the shared data carrier as input through the agent, generate the processing result based on the reference information, and write the processing result into the data area in the shared data carrier corresponding to the system processing stage; wherein, the shared data carrier includes multiple data areas divided according to the system processing stage, and the shared data carrier is used to transfer task processing information between multiple agents; The response processing module 804 is used to send processing results to the user through the interactive interface, respond to the received user response information, parse the user response information, and update the system processing stage when the user response information is confirmation.

[0094] The multi-Agent collaborative orchestration system for system development provided in this application adopts the multi-Agent collaborative orchestration method for system development in the above embodiments, achieving the effects of automatic cross-stage information inheritance, improved accuracy and compliance of generated results, reduced reliance on manual scheduling, and traceable user confirmation status. Compared with the prior art, the beneficial effects of the multi-Agent collaborative orchestration system for system development provided in this application are the same as those of the multi-Agent collaborative orchestration method for system development provided in the above embodiments, and other technical features of the multi-Agent collaborative orchestration system for system development are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0095] The following is for reference. Figure 9 It shows a schematic diagram of a multi-Agent collaborative orchestration system suitable for implementing the embodiments of this application in system development.

[0096] The system includes: User access layer 910 is the unified interaction entry point between the system and external users and systems. It is used to receive input signals from multiple channels and convert the input signals into standard events that the system can process for the orchestration control layer 920 to process.

[0097] The orchestration control layer 920 is responsible for event routing, task orchestration, context management, and state transitions. The orchestration control layer 920 includes an engineering layer gateway 921, an intent-recognizing agent 922, and a Research Context Object (RCO) global manager 923. The engineering layer gateway 921, intent-recognizing agent 922, and RCO global manager 923 work together to transform external events into tasks executable by the corresponding agents in the AI ​​intelligent brain layer 930, and manage the state transitions throughout the task's lifecycle.

[0098] The AI ​​intelligent brain layer 930 is the core of the system's business execution, including but not limited to the demand agent 931, the testing agent 932, the collaborative scheduling agent 933, and the early warning agent 934. Each business agent is an independent agent with large language model reasoning ability, RAG knowledge layer 950 knowledge enhancement ability, and Model Context Protocol (MCP) tool invocation ability, and cross-agent information inheritance is realized through the RCO global manager 923.

[0099] Capability Layer 940 uses MCP standardization to access external tool capabilities, providing a unified tool invocation interface for each intelligent agent in the AI ​​intelligent brain layer 930. Capability Layer 940 is used to standardize, audit, and controllably manage tool invocations, enabling each intelligent agent to invoke heterogeneous external systems in a consistent manner and ensuring that every tool invocation is traceable and controllable.

[0100] RAG Knowledge Layer 950 is the system's knowledge infrastructure, including but not limited to a compliance rule base 951, a requirements clarification knowledge base 952, a test point recall knowledge base 953, a historical use case knowledge base 954, and a graph database 955. It provides semantically retrieved contextual knowledge injection services to each agent and receives knowledge feedback data from each agent to achieve continuous self-optimization of the knowledge base. Through a structured knowledge base design and a closed-loop knowledge feedback mechanism, RAG Knowledge Layer 950 transforms individual experience into organizational-level capability assets to address the problem of domain knowledge gaps.

[0101] This application provides a multi-agent collaborative orchestration device for system development, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to execute the multi-agent collaborative orchestration method for system development described in Embodiment 1 above.

[0102] The following is for reference. Figure 10 This document illustrates a structural schematic diagram of a multi-agent collaborative orchestration device suitable for implementing system development embodiments of this application. The multi-agent collaborative orchestration device for system development in this application embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 10 The multi-agent collaborative orchestration device shown for system development is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0103] like Figure 10As shown, a multi-agent collaborative orchestration device for system development may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the multi-agent collaborative orchestration device for system development. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows a multi-agent collaborative orchestration device for system development to communicate wirelessly or wiredly with other devices to exchange data. While the figure shows a multi-agent collaborative orchestration device for system development with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0104] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0105] The multi-Agent collaborative orchestration device for system development provided in this application adopts the multi-Agent collaborative orchestration method for system development in the above embodiments, achieving the effects of automatic cross-stage information inheritance, improved accuracy and compliance of generated results, reduced reliance on manual scheduling, and traceable user confirmation status. Compared with the prior art, the beneficial effects of the multi-Agent collaborative orchestration device for system development provided in this application are the same as those of the multi-Agent collaborative orchestration method for system development provided in the above embodiments, and other technical features in this multi-Agent collaborative orchestration device for system development are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0106] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0107] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0108] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the multi-Agent collaborative orchestration method for system development in the above embodiments.

[0109] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0110] The aforementioned computer-readable storage medium may be included in a multi-agent collaborative orchestration device for system development; or it may exist independently and not be assembled into a multi-agent collaborative orchestration device for system development.

[0111] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by a multi-agent collaborative orchestration device developed for system development, the device transmits task processing information among multiple agents through a shared data carrier. This eliminates the manual information transmission link, reduces compliance risks of generated results by actively injecting reference information, and achieves automatic state transition for human-machine collaborative confirmation by responding to user replies. It transforms user confirmation into a traceable system state, achieving the effects of automatic cross-stage information inheritance, improved accuracy and compliance of generated results, reduced reliance on manual scheduling, and traceable user confirmation status.

[0112] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0113] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0114] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0115] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., computer programs) for executing the aforementioned multi-Agent collaborative orchestration method for system development. This achieves the effects of automatic cross-stage information inheritance, improved accuracy and compliance of generated results, reduced reliance on manual scheduling, and traceable user confirmation status. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the multi-Agent collaborative orchestration method for system development provided in the above embodiments, and will not be elaborated upon here.

Claims

1. A multi-agent collaborative orchestration method for system development, characterized in that, The method includes: In response to a received user request, intent recognition is performed on the user request to determine the target task and at least one of a plurality of intelligent agents that handle the target task; Based on the key information of the target task, reference information is retrieved from at least one knowledge base and injected into the intelligent agent; The intelligent agent reads the task processing information in the shared data carrier as input, generates a processing result based on the reference information, and writes the processing result into the data area in the shared data carrier corresponding to the system processing stage; wherein, the shared data carrier includes multiple data areas divided according to the system processing stage, and the shared data carrier is used to transmit the task processing information among the multiple intelligent agents; The system sends the processing result to the user through an interactive interface, and in response to the received user reply, it parses the user reply. When the user reply is confirmed, the system updates the processing stage.

2. The multi-agent collaborative orchestration method for system development as described in claim 1, characterized in that, The step of performing intent recognition on the user request to determine the target task and at least one of a plurality of intelligent agents processing the target task includes: Based on the semantics of the user request, identify the target task corresponding to the user request; Based on the task type of the target task, at least one agent is determined from the plurality of agents to handle the target task.

3. The multi-agent collaborative orchestration method for system development as described in claim 1, characterized in that, The step of retrieving reference information from at least one knowledge base based on the key information of the target task includes: Extract key information of the target task, convert the key information into a query vector, perform semantic retrieval on the at least one knowledge base, and obtain multiple candidate knowledge fragments; The candidate knowledge fragments are sorted and selected based on their relevance. The selected knowledge fragments are combined with the key information of the target task to obtain the reference information.

4. The multi-agent collaborative orchestration method for system development as described in claim 1, characterized in that, The step of reading task processing information from the shared data carrier as input by an intelligent agent and generating processing results based on the reference information includes: The intelligent agent determines the data region in the shared data carrier corresponding to the current system processing stage based on the current system processing stage, and obtains the task processing information. The processing result is generated by inputting the task processing information and the injected reference information into the intelligent agent.

5. The multi-agent collaborative orchestration method for system development as described in claim 1, characterized in that, Following the response to the received user reply, the following is also included: Perform semantic parsing on the user response information to identify the type of the user response information; When the response information is modified, the data area in the shared data carrier corresponding to the current processing stage is updated based on the modification opinion, and the agent is triggered to regenerate the processing result based on the modification opinion; When the response is a rejection, the agent is triggered to re-execute the processing flow corresponding to the target task.

6. The multi-agent collaborative orchestration method for system development as described in claim 1, characterized in that, The method further includes: Heterogeneous signals collected from multiple heterogeneous signal sources are normalized, and a comprehensive risk score is obtained based on the information already written in the shared data carrier and the normalized heterogeneous signals. The comprehensive risk score triggers a tiered early warning system.

7. The multi-agent collaborative orchestration method for system development as described in claim 1, characterized in that, Following the update of the system processing phase, the following is also included: After the processing results generated by the intelligent agent are structured, they are asynchronously written into the at least one knowledge base; The retrieval weights of information in at least one knowledge base are adjusted based on the adoption of the reference information.

8. A multi-agent collaborative orchestration system for system development, characterized in that, The system includes: An intent recognition module is used to, in response to a received user request, recognize the intent of the user request to determine the target task and at least one of a plurality of intelligent agents that process the target task. A knowledge retrieval module is used to retrieve reference information from at least one knowledge base based on key information of the target task, and inject the reference information into the intelligent agent; The intelligent agent execution module is used to read task processing information from the shared data carrier as input through the intelligent agent, generate processing results based on the reference information, and write the processing results into the data area in the shared data carrier corresponding to the system processing stage; wherein, the shared data carrier includes multiple data areas divided according to the system processing stage, and the shared data carrier is used to transmit the task processing information among the multiple intelligent agents; The response processing module is used to send the processing result to the user through an interactive interface, parse the user response information in response to the received user response information, and update the system processing stage when the user response information is confirmation.

9. A multi-agent collaborative orchestration device for system development, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the multi-agent collaborative orchestration method for system development as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the multi-Agent collaborative orchestration method for system development as described in any one of claims 1 to 7.

Images