A Zero-Code AI Agent Development and Deployment Data Management System and Method

Abstract

The present invention relates to the field of agent development, and specifically to a zero-code AI agent development and deployment data management system and method, including: a data flow testing module, a model connectivity module, a feedback processing module, a project migration module, and a secondary development module. The data flow testing module is used to perform connectivity testing and obtain conversion interfaces. The model connectivity module is used to connect to the conversion interfaces in the forward direction and generate a conversion queue. The feedback processing module is used to share input data flows between models. The project migration module is used to adjust the shared data ratio of the models. The secondary development module is used to develop new models according to the service provision ratio of the models. The present invention can generate more accurate output content by sharing data, continuously optimize the structure of the agent, obtain a good performance optimization curve during the service process of the agent, improve the ability to process data, and enhance the operation efficiency of the agent system.

Description

Technical Field

[0001] The present invention relates to the field of agent development, and specifically to a zero-code AI agent development and deployment data management system and method. Background Technique

[0002] Zero-code AI agent development refers to the technology of quickly constructing available AI agents through model recognition configuration and customization operations without writing complex codes by using ready-made AI development platforms and tools. Agents deployed through zero code can perform tasks such as natural language conversations, image recognition, and predictive analysis.

[0003] Zero-code agents achieve interaction with users through the processing of data streams by intelligent models. Therefore, for the development of zero-code agents, matching and fitting for big data models are required. However, limited by the functions of the development platform, each model is independent of each other, which limits innovation and customization during the process of developers developing agents with zero code, and enables users to only choose one to use in the interaction model with the agent, making it impossible to meet some complex or personalized requirements.

[0004] In addition, after deploying the model, developers also need to develop new models according to user feedback. However, most users do not understand the working principle of the model and can only infer the user's requirements for the model functions through the user's usage process, which requires additional investigation work by developers and is not conducive to the secondary development of the agent model by developers. Summary of the Invention

[0005] The purpose of the present invention is to provide a zero-code AI agent development and deployment data management system and method to solve the problems raised in the above background technique.

[0006] To solve the above technical problems, the present invention provides the following technical solution: A zero-code AI agent development and deployment data management system, including: a data stream testing module, a model connection module, a feedback processing module, a project migration module, and a secondary development module;

[0007] The data stream testing module is used to generate random vector data streams, perform data connectivity tests on each call interface between development models, use the data stream corresponding to the highest data connectivity as the characteristic data stream, and test the latency of the call interfaces between models, and use the interface with the lowest latency as the conversion interface between models;

[0008] The model connection module is used to calculate the consistency of the characteristic data streams of each model, arrange all development models in descending order of consistency, connect the conversion interfaces between models in the forward direction to form a model conversion queue, develop a supporting processing program based on the model conversion queue, and deploy the program on the service platform;

[0009] The feedback processing module is used to obtain the input data stream provided by the user, calculate the consistency between the input data stream and each model feature data stream, select a processing model from the model queue in the order of consistency, point the directional pointer to the model with the highest consistency, make the sum of the communication delays of the model - to - model conversion interfaces less than a preset threshold, share the input data stream among the processing models according to the consistency ratio, and output a reply data stream to the user on the service platform after model processing;

[0010] The project migration module is used to obtain the feedback data stream of the user. When the consistency between the feedback data stream and the standard forward data stream is less than the consistency between two model feature data streams, increase the shared data ratio of the subsequent model in the model conversion queue by one unit; otherwise, decrease the shared data ratio by one unit, and re - obtain the feedback data stream of the user until the shared data ratio of the previous model drops to the threshold, then close the conversion interface of the previous model and move the directional pointer one position;

[0011] The secondary development module is used to output the position of the directional pointer and the data sharing ratio of each model in the model conversion queue during the service process after the user goes offline. Multiply the sharing ratio, sharing duration, and time decay coefficient to obtain the service weight of each model. After the development cycle ends, develop a new model according to the service weight of the model, and deploy the agent to the service platform again.

[0012] Further, the data stream testing module includes: a connectivity testing unit and a latency testing unit;

[0013] The connectivity testing unit is used to generate a random vector data stream, test the connectivity between every two models, and store the data stream generated when the connectivity is the highest;

[0014] The latency testing unit is used to test the call latency of each interface between models and mark the conversion interface with the lowest latency.

[0015] Further, the model connectivity module includes: a consistent arrangement unit and an interface connection unit;

[0016] The consistent arrangement unit is used to calculate the consistency of the feature data streams between models and arrange all models in the order of consistency;

[0017] The interface connection unit is used to unidirectionally connect all models through the conversion interfaces between models to form a model conversion queue and set the initial pointer of the queue.

[0018] Further, the feedback processing module includes: an input - output unit, a model deployment unit, and a ratio access unit;

[0019] The input / output unit is used to obtain the input content of the user on the agent platform, convert it into an input data stream and upload it to the server, and convert the reply data stream fed back by the server into natural language output;

[0020] The model deployment unit is used to deploy the model conversion queue in the server, receive the input data stream, and output the reply data stream;

[0021] The proportional access unit is used to select a model from the model conversion queue according to the consistency between the input data stream and the feature data stream, and receive the reply data of the model in proportion.

[0022] Further, the project migration module includes: a pointer sliding unit, a model conversion unit, and a proportional scaling unit;

[0023] The pointer sliding unit is used to judge the reply effect of the previous reply data according to the feedback data stream of the user, and convert the reply model according to the level of the reply effect;

[0024] The model conversion unit is used to move the directional pointer, adjust the data sharing ratio of the reply model, and turn on or off the model on the server;

[0025] The proportional scaling unit is used to calculate the adjustment amount of the sharing ratio, so that the model selection method can quickly match the user requirements.

[0026] Further, the secondary development module includes: a weight output unit and a redeployment unit;

[0027] The weight output unit is used to output the service weights of each model during the service process after the user logs off from the agent platform;

[0028] The redeployment unit is used to give the development direction of the new model during the development cycle and redeploy the new model to the agent platform.

[0029] A zero-code AI agent development and deployment data management method includes the following steps:

[0030] Step S1. Select the interface with the lowest latency between development models as the conversion interface, generate a random data stream, perform a data connectivity test on the conversion interface of the development model, and record the data stream generated when the connectivity is the highest as the feature data stream of the model;

[0031] Step S2. Use the data stream comparison method to obtain the consistency of the feature data streams of each model, arrange all the development models in descending order of consistency, and connect the models into a queue through the conversion interface to form a model conversion queue and deploy it in the server;

[0032] Step S3. The user provides an input data stream to the agent, calculates the consistency between the input data stream and each model feature data stream, selects the model with the highest consistency as the starting model, and shares the input data among the starting model and the subsequent models in the starting model queue according to the consistency ratio, and obtains the output data after processing;

[0033] Step S4. The user provides a feedback data stream for the output data, adjusts the sharing ratio of the input data of each model in the model conversion queue according to the consistency between the current feedback data stream and the historical feedback data stream, and repeats the execution until the sharing ratio of the starting model is lower than the threshold, and replaces the starting model in the server;

[0034] Step S5. After the user goes offline, accumulate the data sharing ratio, sharing duration, and time decay coefficient of each model during the service process to obtain the service weight of the model. After the development cycle ends, output the ratio of all service weights of the model during the development cycle, and redeploy the model.

[0035] Further, step S1 includes:

[0036] Step S11. Perform a ping command test on all interfaces of the developed model to obtain the feedback latency of the interfaces, and use the interface with the minimum latency in the model as the conversion interface of the model. The conversion interface is used for signal transmission and data sharing between models;

[0037] Step S12. Connect every two models through the conversion interface, use natural language generation software to generate a random data stream suitable for the agent. The data volume of the random data stream is a fixed value, input the random data stream into the first model, and obtain the output from the second model, and calculate the ratio between the data volume of the output data and the data volume of the input data, which is recorded as the connectivity of the first model;

[0038] Step S13. Repeat generating the random data stream a times, where a is the preset number of test times, record the random data stream generated when the connectivity of each model is the highest during the test process, and use this random data stream as the feature data stream of the model.

[0039] Further, step S2 includes:

[0040] Step S21. Calculate the consistency of each model feature data stream according to the data stream comparison method. The data stream comparison method is as follows:

[0041]

[0042] Among them, Q represents the consistency of the current model feature data stream, n represents the number of models, m represents the data volume of the feature data stream, C1 j and C1 j+1 represent the jth and (j + 1)th data points in the current model feature data stream, Cij and Ci j+1 represent the j-th and (j + 1)-th data points in the feature data stream of the i-th model;

[0043] Step S22. Arrange all models in descending order of the consistency coefficient, and linearly connect the models in the arranged order through the conversion interfaces in the models to form a model conversion queue;

[0044] Step S23. Deploy the model conversion queue in the server, obtain the input content of the user on the front-end intelligent agent platform, convert it into an input data stream and upload it to the server, and after the server finishes processing the input data, output the output data stream fed back by the server through the intelligent agent platform.

[0045] Further, step S3 includes:

[0046] Step S31. The user interacts with the intelligent agent, and the front-end of the intelligent agent converts the user's interactive input into a data stream format and sends it to the server as the user's input data stream, calculates the consistency between the input data stream and the feature data streams of each model, and the model with the highest consistency is used as the starting model;

[0047] Step S32. Set a pointer pointing to the starting model in the queue, and select the c models after the starting model in the queue as shared models, where c is the maximum value that satisfies the inequality T1 + T2 + … + Tc < TR, where T1, T2, … Tc respectively represent the communication delays between the 1st to c-th models after the starting model and the subsequent adjacent models, and TR is a preset delay threshold;

[0048] Step S33. Calculate the input data sharing ratio yc of each shared model, where yc = Qc / ΣQ, where Qc represents the consistency between the feature data stream of the shared model and the input data stream, and ΣQ represents the sum of the consistencies between the feature data streams of all models and the input data stream;

[0049] Step S34. Split the input data according to the ratio and send it to the shared models for processing respectively, and use the Cluster data integration tool to summarize the output data processed by each model and send it to the front-end intelligent agent as feedback data.

[0050] Further, step S4 includes:

[0051] Step S41. The front-end intelligent agent obtains the feedback data of the user and adjusts the data sharing ratio according to the following method:

[0052] S41-1. If there is no historical feedback data, go to S41-2; if there is historical feedback data, go to S41-3;

[0053] S41-2. Calculate the consistency between the feedback data stream and the standard forward data stream. If it is less than the consistency between the starting model and the subsequent model in the queue, reduce the data sharing ratio of the starting model by a preset unit, and distribute the reduced ratio among the remaining shared models. Otherwise, increase the data sharing ratio of the starting model by a preset unit, and the increased ratio is provided by the remaining shared models;

[0054] S41-3. Calculate the consistency between the current feedback data stream and the previous feedback data stream. If it is less than the consistency between the starting model and the subsequent model in the queue, reduce the data sharing ratio of the starting model by u units, where u = (QB - QN) / QE, QB represents the consistency between the current feedback data stream and the previous feedback data stream, QN represents the average value of the consistency between historical feedback data streams, and QE represents the consistency between the starting model and the subsequent model in the queue. Otherwise, increase the data sharing ratio of the starting model by u units;

[0055] Step S42. Execute step S41 once every time the user's feedback data stream is obtained. If the data sharing ratio of the starting model is lower than the preset threshold after execution, mark the model with the highest data sharing ratio among the remaining shared models as the new starting model.

[0056] Further, step S5 includes:

[0057] Step S51. After the interaction between the user and the agent ends, calculate the service weights of each model:

[0058]

[0059] Among them, Rc represents the service weight of the model, k represents the number of user interactions, yv represents the data sharing ratio of the model during the vth interaction, Tv represents the duration of the vth user interaction, and h represents the preset time decay coefficient;

[0060] Step S52. After the development cycle ends, calculate the sum of the service weights of each model during the development cycle, develop new models according to the proportion of the calculation results, and redeploy them in the agent server.

[0061] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0062] 1. The present invention can use random data streams to perform data connectivity tests on various models, use the data stream corresponding to the highest data connectivity as the characteristic data stream, arrange the models according to the vector consistency of the characteristic data streams of each model, connect the conversion interfaces in the forward direction, make isolated models communicate with each other, generate more accurate output content by sharing data, can improve the data processing ability, and enhance the accuracy, robustness and operation efficiency of the agent system.

[0063] 2. The present invention can analyze the consistency between the input data stream and the feature data stream according to the data stream provided by the user to the intelligent agent, share the input data stream among models through the conversion interface according to the consistency ratio, adjust the type and call ratio of the models according to the user's preference, improve the user experience, and continuously optimize the output content of the intelligent agent, so as to obtain a good performance optimization curve during the service process of the intelligent agent.

[0064] 3. After the user goes offline, the present invention outputs the connection weights of the models according to the current model access ratio. In each development cycle, the model with the highest connection weight is developed into a new model, eliminating the need for user research, which can accelerate the development progress of the intelligent agent, enhance the humanized interaction of the intelligent agent, and ensure high competitiveness and high efficiency in the development process. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0066] Figure 1 is a schematic structural diagram of a zero-code AI intelligent agent development and deployment data management system of the present invention;

[0067] Figure 2 is a schematic step diagram of a zero-code AI intelligent agent development and deployment data management method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0068] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0069] Please refer to Figure 1 , the present invention provides a technical solution: a zero-code AI intelligent agent development and deployment data management system, including: a data stream testing module, a model connectivity module, a feedback processing module, a project migration module, and a secondary development module;

[0070] The data stream testing module is used to generate a random vector data stream, perform data connectivity testing on each call interface between development models, use the data stream corresponding to the highest data connectivity as the feature data stream, and test the latency of the call interfaces between models, and use the interface with the lowest latency as the conversion interface between models;

[0071] The data stream testing module includes: a connectivity testing unit and a latency testing unit;

[0072] The connectivity test unit is used to generate a random vector data stream, test the connectivity between every two models, and store the data stream generated when the connectivity is the highest.

[0073] The latency test unit is used to test the call latency of each interface between models and mark the conversion interface with the lowest latency.

[0074] The model connectivity module is used to calculate the consistency of the characteristic data streams of each model, arrange all development models in descending order of consistency, connect the conversion interfaces between models in the forward direction to form a model conversion queue, develop a supporting processing program based on the model conversion queue, and deploy the program on the service platform.

[0075] The model connectivity module includes: a consistent arrangement unit and an interface connection unit.

[0076] The consistent arrangement unit is used to calculate the consistency of the characteristic data streams between models and arrange all models in the order of consistency.

[0077] The interface connection unit is used to unidirectionally connect all models through the conversion interfaces between models to form a model conversion queue and set the initial pointer of the queue.

[0078] The feedback processing module is used to obtain the input data stream provided by the user, calculate the consistency between the input data stream and the characteristic data streams of each model, select a processing model on the model queue in the order of consistency, point the directional pointer to the model with the greatest consistency, make the sum of the communication latencies of the conversion interfaces between models less than a preset threshold, share the input data stream in proportion in the processing model, and output a reply data stream to the user on the service platform after model processing.

[0079] The feedback processing module includes: an input / output unit, a model deployment unit, and a proportional access unit.

[0080] The input / output unit is used to obtain the input content of the user on the intelligent agent platform, convert it into an input data stream and upload it to the server, and convert the reply data stream fed back by the server into natural language for output.

[0081] The model deployment unit is used to deploy the model conversion queue in the server, receive the input data stream, and output a reply data stream.

[0082] The proportional access unit is used to select a model on the model conversion queue according to the consistency between the input data stream and the characteristic data stream and receive the reply data of the model in proportion.

[0083] The project migration module is used to obtain the feedback data stream of the user. When the consistency between the feedback data stream and the standard forward data stream is less than the consistency between two model feature data streams, the shared data ratio of the subsequent model in the model conversion queue is increased by one unit; otherwise, the shared data ratio is decreased by one unit, and the feedback data stream of the user is obtained again until the shared data ratio of the previous model drops to the threshold, then the conversion interface of the previous model is closed and the directional pointer is moved one bit.

[0084] The project migration module includes: a pointer sliding unit, a model conversion unit, and a ratio scaling unit;

[0085] The pointer sliding unit is used to judge the reply effect of the previous reply data according to the feedback data stream of the user, and convert the reply model according to the level of the reply effect;

[0086] The model conversion unit is used to move the directional pointer, adjust the data sharing ratio of the reply model, and turn on or off the model on the server;

[0087] The ratio scaling unit is used to calculate the adjustment amount of the sharing ratio, so that the model selection method can quickly match the user's needs.

[0088] The secondary development module is used to output the position of the directional pointer and the data sharing ratio of each model in the model conversion queue during the service process after the user goes offline. By multiplying the sharing ratio, sharing duration, and time decay coefficient, the service weight of each model is obtained. After the development cycle ends, a new model is developed according to the service weight of the model, and the agent is deployed to the service platform again.

[0089] The secondary development module includes: a weight output unit and a redeployment unit;

[0090] The weight output unit is used to output the service weight of each model during the service process after the user logs off from the agent platform;

[0091] The redeployment unit is used to give the development direction of the new model during the development cycle and redeploy the new model to the agent platform.

[0092] As Figure 2 shown, a zero-code AI agent development and deployment data management method includes the following steps:

[0093] Step S1. Select the interface with the lowest latency between development models as the conversion interface, generate a random data stream, perform a data connectivity test on the conversion interface of the development model, and the data stream generated when the connectivity is the highest is recorded as the feature data stream of the model;

[0094] Step S1 includes:

[0095] Step S11. Perform a ping command test on all interfaces of the development model to obtain the feedback latency of the interfaces. Select the interface with the minimum latency in the model as the conversion interface of the model, and the conversion interface is used for signal transmission and data sharing between models.

[0096] Step S12. Connect every two models through the conversion interface, use natural language generation software to generate a random data stream suitable for the agent. The data volume of the random data stream is a fixed value. Input the random data stream into the first model and obtain the output from the second model. Calculate the ratio between the data volume of the output data and the data volume of the input data, which is denoted as the connectivity of the first model.

[0097] Step S13. Repeat generating the random data stream a times, where a is the preset number of test times. Record the random data stream generated when the connectivity of each model is the highest during the test process, and use this random data stream as the characteristic data stream of the model.

[0098] Step S2. Use the data stream comparison method to obtain the consistency of the characteristic data streams of each model. Arrange all development models in descending order of consistency, and connect the models in a queue through the conversion interface to form a model conversion queue and deploy it in the server.

[0099] Step S2 includes:

[0100] Step S21. Calculate the consistency of the characteristic data streams of each model according to the data stream comparison method. The data stream comparison method is as follows:

[0101]

[0102] where Q represents the consistency of the characteristic data stream of the current model, n represents the number of models, m represents the data volume of the characteristic data stream, C1 j and C1 j+1 represent the jth and (j + 1)th data points in the characteristic data stream of the current model, and Ci j and Ci j+1 represent the jth and (j + 1)th data points in the characteristic data stream of the ith model;

[0103] Step S22. Arrange all models in descending order of the consistency coefficient, and linearly connect the models through the conversion interfaces in the models according to the arrangement order to form a model conversion queue;

[0104] Step S23. Deploy the model conversion queue in the server, obtain the input content of the user on the front-end agent platform, convert it into an input data stream and upload it to the server, and after the server finishes processing the input data, output the output data stream fed back by the server through the agent platform.

[0105] Step S3. The user provides an input data stream to the agent, calculates the consistency between the input data stream and each model feature data stream, selects the model with the highest consistency as the starting model, and shares the input data among the starting model and the subsequent models in the starting model queue according to the consistency ratio, and the output data is obtained after processing.

[0106] Step S3 includes:

[0107] Step S31. The user interacts with the agent, and the front-end of the agent converts the user's interactive input into a data stream format and sends it to the server as the user's input data stream, calculates the consistency between the input data stream and each model feature data stream, and selects the model with the highest consistency as the starting model.

[0108] Step S32. Set a pointer pointing to the starting model in the queue, select the c subsequent models of the starting model in the queue as the shared models, where c is the maximum value that satisfies the inequality T1 + T2 + … + Tc < TR, where T1, T2, … Tc represent the communication delays between the 1st to cth models after the starting model and their subsequent adjacent models respectively, and TR is a preset delay threshold.

[0109] Step S33. Calculate the input data sharing ratio yc of each shared model, where yc = Qc / ΣQ, Qc represents the consistency between the shared model feature data stream and the input data stream, and ΣQ represents the sum of the consistencies between all model feature data streams and the input data stream.

[0110] Step S34. Split the input data according to the ratio and send it to the shared models for processing respectively, and use the Cluster data integration tool to summarize the output data after processing by each model and send it to the front-end agent as the feedback data.

[0111] Step S4. The user provides a feedback data stream for the output data, adjusts the sharing ratio of the input data of each model in the model conversion queue according to the consistency between the current feedback data stream and the historical feedback data stream, and repeats the execution until the sharing ratio of the starting model is lower than the threshold, and then replaces the starting model in the server.

[0112] Step S4 includes:

[0113] Step S41. The front-end agent obtains the user's feedback data and adjusts the data sharing ratio according to the following method:

[0114] S41-1. If there is no historical feedback data, go to S41-2; if there is historical feedback data, go to S41-3.

[0115] S41-2. Calculate the consistency between the feedback data stream and the standard forward data stream. If it is less than the consistency between the starting model and the subsequent models in the queue, reduce the data sharing ratio of the starting model by a preset unit, and distribute the reduced ratio among the remaining sharing models. Otherwise, increase the data sharing ratio of the starting model by a preset unit, and the increased ratio is provided by the remaining sharing models;

[0116] S41-3. Calculate the consistency between the current feedback data stream and the previous feedback data stream. If it is less than the consistency between the starting model and the subsequent models in the queue, reduce the data sharing ratio of the starting model by u units, where u = (QB - QN) / QE, QB represents the consistency between the current feedback data stream and the previous feedback data stream, QN represents the average value of the consistency between historical feedback data streams, and QE represents the consistency between the starting model and the subsequent models in the queue. Otherwise, increase the data sharing ratio of the starting model by u units;

[0117] Step S42. Each time a user's feedback data stream is obtained, execute Step S41 once. If the data sharing ratio of the starting model is lower than the preset threshold after execution, mark the model with the highest data sharing ratio among the remaining sharing models as the new starting model.

[0118] Step S5. After the user logs off, accumulate the data sharing ratio, sharing duration, and time decay coefficient of each model during the service process to obtain the service weight of the model. After the development cycle ends, output the ratio of all service weights of the model during the development cycle and redeploy the model.

[0119] Step S5 includes:

[0120] Step S51. After the interaction between the user and the agent ends, calculate the service weight of each model:

[0121]

[0122] Among them, Rc represents the service weight of the model, k represents the number of user interactions, yv represents the data sharing ratio of the model during the vth interaction, Tv represents the duration of the vth user interaction, and h represents the preset time decay coefficient;

[0123] Step S52. After the development cycle ends, calculate the sum of the service weights of each model during the development cycle, develop a new model according to the ratio of the calculation results, and redeploy it in the agent server.

[0124] Example: There are 3 models in the agent server, and the characteristic data streams are [2, 4, 1, 3, 5], [1, 4, 2, 2, 1], and [5, 3, 3, 1, 4] respectively, and the consistencies are 0.18, 0.06, and 0.23 respectively. Then, they are arranged in the order of model 3, model 1, and model 2 to form a model queue, where model 3 is the starting model; during the user's interaction with the agent, the sharing ratios obtained according to the provided input data stream are 0.5, 0.3, and 0.2. After the user's feedback, the consistency of the feedback data stream decreases, and the sharing ratios are reallocated to 0.4, 0.36, and 0.24.

[0125] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0126] Finally, it should be noted that the above are only preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A zero-code AI agent development and deployment data management method, characterized in that: The method comprises the following steps: Step S1. Select the interface with the lowest latency between development models as the conversion interface, generate a random data stream, and perform a data connectivity test on the conversion interface of the development model. The data stream generated when the connectivity is the highest is recorded as the characteristic data stream of the model; Step S2. Use the data stream comparison method to obtain the consistency of the feature data streams of each model, arrange all development models in descending order of consistency, connect the models as queues through the conversion interface, and form a model conversion queue deployed in the server; Step S3. The user provides an input data stream to the agent, calculates the consistency of the input data stream with the feature data streams of each model, and uses the model with the highest consistency as the starting model. The input data is shared among the starting model and the subsequent models in the starting model queue according to the consistency ratio, and the output data is obtained after processing. Step S4. The user provides a feedback data stream for the output data. Based on the consistency between the current feedback data stream and the historical feedback data stream, the sharing ratio of the input data of each model in the model conversion queue is adjusted. The process is repeated until the sharing ratio of the starting model is lower than the threshold, and the starting model is replaced in the server. Step S5. After the user goes offline, the data sharing ratio, sharing duration and time decay coefficient of each model in the service process are accumulated to obtain the service weight of the model. After the development cycle is over, the ratio of the total service weight of the model in the development cycle is output, and the model is redeployed; Step S2 includes: Step S21. The data stream comparison method is as follows: Among them, Q represents the consistency of the current model feature data stream, n represents the number of models, m represents the amount of data in the feature data stream, and C1 j and C1 j+1 represents the jth and j+1th data points in the current model feature data stream, Ci j Ci j+1 Represents the jth and j+1th data points in the feature data stream of the i-th model; Step S22. Arrange the order in descending order of consistency to form a model conversion queue; Step S23. The user's input content is obtained on the front-end agent platform, converted into an input data stream and uploaded to the server, and after the server has processed the input data, the output data stream fed back by the server is output through the agent platform; In step S5, the method for calculating the service weight of each model is as follows: Among them, Rc represents the service weight of the model, k represents the number of user interactions, yv represents the data sharing ratio of the model at the vth interaction, Tv represents the duration of the user's vth interaction, and h represents the preset time decay coefficient. The new model is developed according to the ratio of the sum of the service weights of each model during the development cycle and redeployed in the intelligent server.

2. A zero-code AI agent development and deployment data management method according to claim 1, characterized in that: Step S1 includes: Step S11. Perform a ping command test on all interfaces of the development model to obtain the feedback delay of the interface, and use the interface with the smallest delay in the model as the conversion interface of the model, which is used for signal transmission and data sharing between models; Step S12. Connect every two models through a conversion interface, use natural language generation software to generate a random data stream suitable for the agent, the data volume of the random data stream is a fixed value, input the random data stream to the first model, and obtain output from the second model, calculate the ratio between the data volume of the output data and the data volume of the input data, and record it as the connectivity of the first model; Step S13. Repeat generating random data streams a times, where a is the preset number of tests, and record the random data streams generated when the connectivity of each model is the highest during the test, and use the random data streams as the characteristic data streams of the model.

3. A zero-code AI agent development and deployment data management method according to claim 2, characterized in that: Step S3 includes: Step S31. The user interacts with the agent, and the agent front-end converts the user's interaction input into a data stream format and sends it to the server as the user's input data stream, calculates the consistency between the input data stream and each model feature data stream, and selects the model with the highest consistency as the starting model; Step S32. Set a pointer to the starting model in the queue, and select the c models after the starting model in the queue as the shared models, where c is the maximum value that satisfies the inequality T1 + T2 + … + Tc < TR, and T1, T2, …, Tc respectively represent the communication delays between the 1st to the cth models after the starting model and their subsequent adjacent models, and TR is the preset delay threshold; Step S33. Calculate the input data sharing ratio yc of each shared model, where yc = Qc / ΣQ, Qc represents the consistency between the shared model feature data stream and the input data stream, and ΣQ represents the sum of the consistencies between all model feature data streams and the input data stream; Step S34. Split the input data according to the ratio and send it to the shared models for processing respectively, and use the Cluster data integration tool to summarize the output data processed by each model and send it to the front-end agent as the feedback data.

4. A zero-code AI agent development and deployment data management method according to claim 3, characterized in that: Step S4 includes: Step S41. The front-end agent obtains the user's feedback data and adjusts the data sharing ratio according to the following method: S41-1. If there is no historical feedback data, go to S41-2; if there is historical feedback data, go to S41-3; S41-2. Calculate the consistency between the feedback data stream and the standard positive data stream. If it is less than the consistency between the starting model and the subsequent model in the queue, reduce the data sharing ratio of the starting model by a preset unit, and distribute the reduced ratio among the remaining shared models; otherwise, increase the data sharing ratio of the starting model by a preset unit, and the increased ratio is provided by the remaining shared models; S41-3. Calculate the consistency between the current feedback data stream and the previous feedback data stream. If it is less than the consistency between the starting model and the subsequent model in the queue, reduce the data sharing ratio of the starting model by u units, where u = (QB - QN) / QE, QB represents the consistency between the current feedback data stream and the previous feedback data stream, QN represents the average value of the consistencies between historical feedback data streams, and QE represents the consistency between the starting model and the subsequent model in the queue; otherwise, increase the data sharing ratio of the starting model by u units; Step S42. Execute Step S41 each time the user's feedback data stream is obtained. If the data sharing ratio of the starting model is lower than the preset threshold after execution, mark the model with the highest data sharing ratio among the remaining shared models as the new starting model.

5. A zero-code AI agent development and deployment data management system, the system executing a zero-code AI agent development and deployment data management method as claimed in claim 1, characterized in that: The system includes the following modules: Data stream testing module, model connectivity module, feedback processing module, project migration module, and secondary development module; The data stream testing module is used to generate random vector data streams, perform data connectivity tests on each call interface between development models, use the data stream corresponding to the highest data connectivity as the feature data stream, and test the delay of the call interfaces between models, and use the interface with the lowest delay as the conversion interface between models; The model connectivity module is used to calculate the consistency of the feature data streams of each model, arrange all development models in descending order of consistency, connect the conversion interfaces between the models in a forward direction, form a model conversion queue, develop a supporting processing program based on the model conversion queue, and deploy the program on the service platform; The feedback processing module is used to obtain the input data stream provided by the user, calculate the consistency of the input data stream with the feature data stream of each model, select the processing model on the model queue according to the consistency order, point the directional pointer to the model with the greatest consistency, make the sum of the communication delays of the conversion interface between models less than a preset threshold, share the input data stream in the processing model according to the consistency ratio, and output the reply data stream to the user on the service platform after model processing; The project migration module is used to obtain the user's feedback data stream. When the consistency between the feedback data stream and the standard forward data stream is less than the consistency between the two model feature data streams, the shared data ratio of the subsequent model in the model conversion queue is increased by one unit, otherwise the shared data ratio is reduced by one unit, and the user's feedback data stream is re-acquired until the shared data ratio of the previous model drops to a threshold value, the conversion interface of the previous model is closed, and the directional pointer moves one position; The secondary development module is used to output the position of the directional pointer and the data sharing ratio of each model in the model conversion queue during the service process after the user goes offline, and obtain the service weight of each model by multiplying the sharing ratio, sharing duration and time decay coefficient. After the development cycle is over, a new model is developed according to the service weight of the model, and the intelligent agent is deployed on the service platform again.

6. A zero-code AI agent development and deployment data management system according to claim 5, characterized in that: The data flow test module includes: a connectivity test unit and a delay test unit; The connectivity test unit is used to generate a random vector data stream, test connectivity between every two models, and store the data stream generated when the connectivity is the highest; The delay test unit is used to test the call delay of each interface between models and mark the conversion interface with the lowest delay; The model connectivity module includes: a consistent arrangement unit and an interface connection unit; The consistent arrangement unit is used to calculate the consistency of the feature data streams between the models and arrange all the models in a consistent order; The interface connection unit is used to unidirectionally connect all models through the conversion interface between models, form a model conversion queue, and set the initial pointer of the queue.

7. A zero-code AI agent development and deployment data management system according to claim 6, characterized in that: The feedback processing module includes: an input and output unit, a model deployment unit and a proportional access unit; The input and output unit is used to obtain the user's input content on the agent platform, convert it into an input data stream and upload it to the server, and convert the reply data stream fed back by the server into a natural language output; The model deployment unit is used to deploy the model conversion queue in the server, receive the input data stream, and output the reply data stream; The proportional access unit is used to select a model on the model conversion queue according to the consistency of the input data stream and the feature data stream, and receive the response data of the model in proportion.

8. A zero-code AI agent development and deployment data management system according to claim 7, characterized in that: The project migration module includes: a pointer sliding unit, a model conversion unit and a proportional expansion unit; The pointer sliding unit is used to judge the reply effect of the previous reply data according to the user's feedback data stream, and convert the reply model according to the reply effect; The model conversion unit is used to move the directional pointer, adjust the data sharing ratio of the reply model, and turn the model on or off on the server; The ratio expansion unit is used to calculate the adjustment amount of the sharing ratio so that the model selection method can be quickly matched with user needs.

9. A zero-code AI agent development and deployment data management system according to claim 8, characterized in that: The secondary development module includes: a weight output unit and a redeployment unit; The weight output unit is used to output the service weights of each model in the service process after the user goes offline from the agent platform; The redeployment unit is used to provide a development direction for the new model during the development cycle and to redeploy the new model to the agent platform.

Images