ARCHITECTURE PROPOSAL DEVICE, INFORMATION PROCESSING METHOD, AND PROGRAM

Abstract

Techniques for generating suitable architectures are provided. [Solution] An architecture proposal device according to one aspect of the present disclosure accepts input of user requirements for an architecture related to an information system, generates a proposed architecture of the information system in a large-scale generative model, generates verification results of the proposed architecture in the large-scale generative model, generates inquiries regarding deficiencies in the large-scale generative model, outputs the generated inquiries to a user, accepts input of answers to the inquiries from the user, updates the user requirements based on the input answers, and generates an updated proposed architecture in the large-scale generative model. The architecture proposal device improves the proposed architecture by repeating the process from verifying to updating the proposed architecture.

Description

[Technical field]

[0001] The present disclosure relates to an architecture proposal device, an information processing method, and a program. [Background technology]

[0002] In recent years, technological developments of large-scale generative models, such as large-scale language models, large-scale visual language models, and large-scale speech models, have progressed. It is known that large-scale generative models can acquire common sense by learning a large amount of data. By using such large-scale generative models, any information can be generated. For example, Patent Document 1 proposes using a large-scale language model to generate program code. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7564601 Summary of the Invention [Problem to be solved by the invention]

[0004] The present inventors have found the following problem. That is, it is assumed that an architecture for an information system is designed using a large-scale generative model. In an architecture for an information system, there are a wide variety of patterns of components, such as module types, specifications, and configuration methods. Therefore, it is difficult to enumerate all requirements for the architecture at the initial stage. If the requirements for the architecture are insufficient, it is difficult to generate an appropriate architecture.

[0005] One aspect of the present disclosure has been made in light of the above circumstances, and one of the objectives is to provide a technique for generating an appropriate architecture. [Means for solving the problem]

[0006] In order to solve the above-mentioned problems, the present disclosure employs the following configurations. Note that the following configurations can be appropriately combined.

[0007] An architecture proposal device according to one aspect of the present disclosure includes a control unit. the control unit is configured to execute the following operations: accepting input of user requirements for an architecture related to the information system from a user; causing the large-scale generative model to generate a proposed architecture of the information system by giving a first prompt including the input user requirements and a proposal instruction; causing the large-scale generative model to generate a verification result of the proposed architecture by giving a second prompt including the generated proposed architecture and a verification instruction; if a deficiency exists in the generated verification result, causing the large-scale generative model to generate a query regarding the deficiency by giving a third prompt including the proposed architecture, the verification result, and an instruction to generate a query; outputting the generated query to the user; accepting input of an answer to the query from the user; updating the user requirements based on the input answer; causing the large-scale generative model to generate an updated proposed architecture by giving a fourth prompt including the updated user requirements, the proposed architecture, the verification result, the query, and an instruction to re-propose; generating a verification result of the proposed architecture until no deficiency exists; generating the query, outputting the query, accepting input of the answer to the query, updating the user requirements, and generating the updated proposed architecture, thereby repeating the above operations; and outputting the improved proposed architecture. It is composed.

[0008] In this configuration, architecture proposal, verification, and inquiries to the user are repeated as an alternating dialogue between the user and the large-scale generative model. In this dialogue, by repeating answers to inquiries, it is possible to promote the refinement of user requirements. This makes it possible to refine the proposed architecture. As a result, it is expected that an appropriate architecture will be generated according to this configuration.

[0009] In the architecture proposal device according to the above aspect, the verification instruction may include a verification list including a plurality of predefined verification items. The deficiency item may be a verification item that is evaluated as having a deficiency in verification using a large-scale generative model among the plurality of verification items included in the verification list. According to this configuration, by defining the verification items in advance, it is possible to expect a guarantee of the verification scope.

[0010] In the architecture proposal device according to the above aspect, the third prompt may further include a skill level of the user in architecture design. With this configuration, it is possible to expect that a query item will be generated according to the skill level of the user.

[0011] In the architecture proposal device according to the above aspect, the user's skill level may be inferred from the user's input. The user's skill level may be reflected in the user's input. Therefore, the skill level can be inferred from the user's input. According to this configuration, by automatically inferring the skill level from the user's input, it is expected to reduce the effort required for the user to answer the skill level.

[0012] In the architecture proposal device according to the above aspect, the control unit may be further configured to cause the large-scale generative model to infer the user's skill level by providing a fifth prompt including a user input and an instruction to infer the skill level. This configuration eliminates the need to prepare a dedicated model for inferring the skill level, and is therefore expected to reduce costs.

[0013] In the architecture proposal device according to the above aspect, the control unit may be further configured to, when the input to the inquiry includes a response that the inquiry is unanswerable, cause the large-scale generative model to generate a modified inquiry by giving a sixth prompt including the inquiry, the user's skill level for architecture design, and a modification instruction, output the modified inquiry to the user, and receive an input of a response to the modified inquiry from the user. Updating the user requirements may be configured by updating the user requirements with the response input to the modified inquiry. According to this configuration, when an inquiry that the user cannot answer is generated, the inquiry can be remade to match the user's skill level, thereby obtaining a possibility of eliciting an appropriate response from the user.

[0014] In the architecture proposal device according to the above aspect, the control unit may be configured to further execute storing the user's input, instructions for the large-scale generative model, and the generation result of the large-scale generative model in chronological order as a dialogue history in association with each other. With this configuration, by using the obtained dialogue history, it is possible to verify after the fact whether the generation process of the proposed architecture is appropriate or not.

[0015] In the architecture proposal device according to the above aspect, the control unit may be further configured to output the generated proposed architecture to a user, and to receive an input from the user specifying a portion of the proposed architecture that is not to be changed. The proposal instruction may include an instruction statement indicating that the designated portion should not be changed. According to this configuration, by accepting the designation of the portion that does not need to be changed and reflecting the designation in the re-proposal instruction, it is possible to prevent the designated portion from being changed when re-proposing the architecture.

[0016] In the architecture proposal device according to the above aspect, the control unit may be configured to further execute outputting the generated proposed architecture to a user and receiving an input from the user specifying a portion of the proposed architecture to be changed. The re-proposal instruction may include an instruction statement indicating that the specified portion is to be changed. With this configuration, it is possible to prompt the user to change the specified portion when re-proposing the architecture.

[0017] In the architecture proposal device according to the above aspect, the control unit may be further configured to cause the large-scale generative model to generate alternatives to the proposed architecture by providing a seventh prompt including the proposed architecture, the verification result, and an instruction to generate an alternative. The third prompt may further include the generated alternative. With this configuration, the generation of the alternative can promote understanding of the proposed architecture. By including the generated alternative in the third prompt, it is expected that the accuracy of generating the query item can be improved.

[0018] In the architecture proposal device according to the above aspect, the control unit may be further configured to cause the large-scale generative model to generate a summary of the proposed architecture by providing an eighth prompt including the proposed architecture and a summary instruction, and to output the generated summary to the user. With this configuration, the output summary can make it easier for the user to understand the proposed architecture.

[0019] In the architecture proposal device according to the above aspect, the instruction to generate a query may include an instruction for generating a query item to which an input format of an answer is restricted. With this configuration, a query item that is easy for a user to answer can be obtained.

[0020] In the architecture proposal device according to the above aspect, the instruction to generate a query may include a directive for listing query items in order of priority. With this configuration, query items that are easy for the user to answer can be obtained.

[0021] In the architecture proposal device according to the above aspect, the instruction to generate a query may include an instruction to generate a query item in a form of asking a user's request. With this configuration, it is expected that the possibility of generating an appropriate query item can be improved.

[0022] In the architecture proposal device according to the above aspect, the control unit may be configured to manage the user requirements, the proposed architecture, the verification results, and the inquiries as structured internal information. According to this configuration, since the internal information is structured, it is easy to construct each prompt. Therefore, it is expected that the implementation of the system (architecture proposal device) can be simplified.

[0023] The embodiment of the present disclosure may not be limited to the architecture proposal device (information processing device). As another aspect of the architecture proposal device according to each of the above aspects, one aspect of the present disclosure may be an information processing method (architecture proposal method) that realizes all or part of each of the above configurations, or may be a program, or may be a storage medium that is readable by a machine such as a computer and that stores such a program. Here, the machine-readable storage medium may be a non-transitory medium that accumulates information such as a program by electrical, magnetic, optical, mechanical, or chemical action. The non-transitory storage medium may include a storage medium (CD, DVD, semiconductor memory, etc.), an auxiliary storage device of a computer, an external storage device connected to a computer, etc.

[0024] For example, an information processing method (architecture proposal method) according to one aspect of the present disclosure may be executed by a computer. The information processing method may include accepting input of user requirements for an architecture related to an information system from a user, causing a large-scale generative model to generate a proposed architecture of the information system by giving a first prompt including the input user requirements and a proposal instruction, causing the large-scale generative model to generate a verification result of the proposed architecture by giving a second prompt including the generated proposed architecture and a verification instruction, and if deficiencies exist in the generated verification result, causing the large-scale generative model to generate a query regarding the deficiency by giving a third prompt including the proposed architecture, the verification result, and an instruction to generate a query, outputting the generated query to a user, accepting input of an answer to the query from the user, updating the user requirements with the input answer, causing the large-scale generative model to generate an updated proposed architecture by giving a fourth prompt including the updated user requirements, the proposed architecture, the verification result, the query, and an instruction to re-propose, generating a verification result of the proposed architecture until no deficiencies exist, generating the query, outputting the query, accepting input of the answer to the query, updating the user requirements, and generating the updated proposed architecture, thereby repeating the steps of improving the proposed architecture, and outputting the improved proposed architecture.

[0025] Also, for example, a program according to one aspect of the present disclosure (architecture proposal program) may be a program for causing a computer to execute an information processing method. The information processing method may include accepting input of user requirements for an architecture related to an information system from a user, causing a large-scale generative model to generate a proposed architecture of the information system by giving a first prompt including the input user requirements and a proposal instruction, causing the large-scale generative model to generate a verification result of the proposed architecture by giving a second prompt including the generated proposed architecture and a verification instruction, and if deficiencies exist in the generated verification result, causing the large-scale generative model to generate a query regarding the deficiency by giving a third prompt including the proposed architecture, the verification result, and an instruction to generate a query, outputting the generated query to a user, accepting input of an answer to the query from the user, updating the user requirements with the input answer, causing the large-scale generative model to generate an updated proposed architecture by giving a fourth prompt including the updated user requirements, the proposed architecture, the verification result, the query, and an instruction to re-propose, generating a verification result of the proposed architecture until no deficiencies exist, generating the query, outputting the query, accepting input of the answer to the query, updating the user requirements, and generating the updated proposed architecture, thereby repeating the steps of improving the proposed architecture, and outputting the improved proposed architecture. Effect of the Invention

[0026] According to one aspect of the present disclosure, a technique for generating an appropriate architecture can be provided. [Brief description of the drawings]

[0027] [Figure 1] FIG. 1 illustrates an example of a situation to which the present disclosure is applied. [Diagram 2]FIG. 2 is a schematic diagram showing an example of a scene in which an input of a user requirement is accepted and a first prompt is given. [Diagram 3] FIG. 3 shows a schematic example of a scene in which the second prompt is given. [Figure 4] FIG. 4 shows a schematic example of a scene in which the third prompt is given. [Diagram 5] FIG. 5 is a schematic diagram showing an example of a scene in which a query is output, an answer is input, and a fourth prompt is given. [Figure 6] FIG. 6 illustrates an example of a method for generating inquiries according to skill levels. [Figure 7] FIG. 7 shows a schematic diagram of an example of a method for modifying a query item. [Figure 8] FIG. 8 illustrates a schematic diagram of one example of a method for controlling the form of a query. [Figure 9] FIG. 9 shows an example of a scene in which an alternative plan is generated. [Figure 10] FIG. 10 illustrates a schematic diagram of one example of a method for controlling the scope of architecture re-proposals. [Figure 11] FIG. 11 shows a schematic diagram of an example of a scene in which a summary is generated. [Figure 12] FIG. 12 shows an example of a scene in which a dialogue history is generated. [Figure 13] FIG. 13 illustrates an example of the internal information. [Figure 14] FIG. 14 shows a specific example of the transition of internal information. [Figure 15] FIG. 15 shows a specific example of the transition of internal information. [Figure 16] FIG. 16 shows a specific example of the transition of internal information. [Figure 17] FIG. 17 shows a specific example of the transition of internal information. [Figure 18] FIG. 18 shows a specific example of the transition of internal information. [Figure 19] FIG. 19 shows a specific example of the transition of internal information. [Figure 20] FIG. 20 shows a specific example of the transition of internal information. [Figure 21] FIG. 21 illustrates an example of a hardware configuration of the architecture proposal device. [Figure 22] FIG. 22 illustrates an example of a software configuration of the architecture proposal device. [Diagram 23] FIG. 23 is a flowchart illustrating an example of a processing procedure for proposing an architecture by the architecture proposal device. [Figure 24] FIG. 24 is a flowchart illustrating an example of a processing procedure for proposing an architecture by the architecture proposal device. [Diagram 25] FIG. 25 shows the architecture originally proposed in the first experimental example. [Figure 26] FIG. 26 shows the architecture originally proposed in the first experimental example. [Figure 27] FIG. 27 shows the verification results of the initially proposed architecture in the first experimental example. [Figure 28] FIG. 28 shows inquiries and answers regarding the architecture originally proposed in the first experimental example. [Figure 29] FIG. 29 shows the architecture proposed for the second time in the first experimental example (the result of the first re-proposal of the architecture). [Diagram 30] FIG. 30 shows the architecture proposed for the second time in the first experimental example (the result of the first re-proposal of the architecture). [Diagram 31] FIG. 31 shows the verification results of the architecture proposed for the second time in the first experimental example. [Diagram 32] FIG. 32 shows inquiries and answers regarding the architecture proposed for the second time in the first experimental example. [Diagram 33] FIG. 33 shows the architecture proposed for the third time in the first experimental example (the result of the second re-proposal of the architecture). [Diagram 34] FIG. 34 shows the architecture proposed for the third time in the first experimental example (the result of the second re-proposal of the architecture). [Diagram 35] FIG. 35 shows instructions given to the large-scale generative model as instructions for generating queries in the form of asking the user's requirements. [Diagram 36] FIG. 36 shows the result of generating a query obtained by giving a prompt that does not include an instruction for generating a query in the form of asking the user's request. [Figure 37] FIG. 37 shows the result of generating a query obtained by giving a prompt including an instruction for generating a query in the form of asking a user's request. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0028] Hereinafter, an embodiment according to one aspect of the present disclosure will be described with reference to the drawings. However, the embodiment described below is merely an example of the present disclosure in all respects. Various improvements or modifications may be made without departing from the scope of the present disclosure. In implementing the present disclosure, a specific configuration according to the embodiment may be appropriately adopted. Note that, although the data appearing in this embodiment is described in natural language, more specifically, it is specified by pseudo-language, commands, parameters, machine language, electrical signals, etc. that can be recognized by a machine such as a computer.

[0029] §1 Examples of application Fig. 1 shows an example of a scene to which the present disclosure is applied. Figs. 2 to 5 show an example of a scene of each step. The architecture proposal device 1 according to this embodiment is one or more computers configured to control a dialogue between a user Z and a large-scale generative model 2, thereby causing the large-scale generative model 2 to construct an architecture related to an information system.

[0030] The architecture proposal device 1 according to this embodiment receives user requirements 30 (30A) for an architecture related to an information system from a user Z (FIG. 1(A) and FIG. 2). The architecture proposal device 1 causes the large-scale generative model 2 to generate a proposed architecture 40 (40A) of the information system by providing a first prompt P1 including the input user requirements 30 (30A) and a proposal instruction 61 (FIG. 1(B) and FIG. 2). The architecture proposal device 1 causes the large-scale generative model 2 to generate a verification result 45 of the proposed architecture 40 (40A) by providing the generated proposed architecture 40 (40A) and a second prompt P2 including a verification instruction 62 (FIG. 1(C) and FIG. 3).

[0031] If there is a deficiency in the generated verification result 45, the architecture proposal device 1 gives a third prompt P3 including the proposed architecture 40 (40A), the verification result 45, and an inquiry generation instruction 63, thereby causing the large-scale generative model 2 to generate an inquiry item 49 regarding the deficiency ((D) of FIG. 1, FIG. 4). The architecture proposal device 1 outputs the generated inquiry item 49 to the user Z, and receives an input of an answer 31 to the inquiry item 49 from the user Z ((E)(F) of FIG. 1, FIG. 5). The architecture proposal device 1 updates the user requirement 30 (30B) with the input answer 31 (FIG. 5). For convenience of explanation, the user requirement 30 before the update is also referred to as the "user requirement 30A", and the user requirement 30 after the update is also referred to as the "user requirement 30B".

[0032] The architecture proposal device 1 causes the large-scale generative model 2 to generate an updated proposed architecture 40 (40B) by providing a fourth prompt P4 including the updated user requirements 30 (30B), the proposed architecture 40 (40A), the verification results 45, the inquiry items 49, and the re-proposal instructions 64 (FIG. 1(G) and FIG. 5). For ease of explanation, the proposed architecture 40 before the update (re-proposal) is also referred to as the "proposed architecture 40A," and the proposed architecture 40 after the update (re-proposal) is also referred to as the "proposed architecture 40B."

[0033] Until no deficiencies exist, the architecture proposal device 1 improves the proposed architecture 40 by repeating the steps of generating verification results 45 of the proposed architecture 40, generating inquiries 49, outputting the inquiries 49, accepting input of answers 31 to the inquiries 49, updating the user requirements 30, and generating the updated proposed architecture 40. The architecture proposal device 1 outputs the improved proposed architecture 40.

[0034] The input of the user requirements 30 and answers 31 may be received in any manner. The matching item 49 may be output by any method. That is, the method of inputting and outputting information to and from the user Z is not particularly limited, and may be appropriately selected according to the embodiment.

[0035] In one example, as shown in each figure, a user Z may access the architecture proposal device 1 using a user terminal U1. The user terminal U1 may be a computer separate from the architecture proposal device 1. The type of computer of the user terminal U1 is not particularly limited and may be appropriately selected depending on the embodiment. The user terminal U1 may be, for example, a general-purpose PC (Personal Computer), a notebook PC, a terminal device, etc. The terminal device is It may include a smartphone, a tablet terminal, etc. In this case, input and output of information to and from the user Z may be performed via the user terminal U1.

[0036] That is, the architecture proposal device 1 may indirectly accept the input of the user requirements 30 and the answers 31 via the user terminal U1. The transmission paths of the input user requirements 30 and the answers 31 may be appropriately selected according to the embodiment. The user requirements 30 and the answers 31 may be transmitted to the large-scale generative model 2 via the architecture proposal device 1, or may be transmitted to the large-scale generative model 2 without passing through the architecture proposal device 1. In addition, the architecture proposal device 1 may output the generated inquiry item 49 to the user terminal U1. The transmission path of the inquiry item 49 may also be appropriately selected according to the embodiment. The inquiry item 49 may be transmitted to the user terminal U1 via the architecture proposal device 1, or may be transmitted to the user terminal U1 without passing through the architecture proposal device 1.

[0037] In another example, the user terminal U1 and the architecture proposal device 1 may be configured as an integrated computer. That is, the architecture proposal device 1 may also operate as the user terminal U1. In this case, the architecture proposal device 1 may directly accept input of the user requirements 30 and answers 31 via an input device. The architecture proposal device 1 may directly output the inquiry items 49 via an output device. The user terminal U1 may be omitted from each diagram.

[0038] In this embodiment, the architecture proposal device 1 repeatedly induces architecture proposal (generation and update of the proposed architecture 40), verification (generation of the verification result 45), and inquiry to the user Z (input of the user requirements 30, output of the inquiry items 49, and reception of the input of the answer 31) as an alternating dialogue between the user Z and the large-scale generative model 2. In this dialogue, the answer 31 to the inquiry item 49 is repeated, so that the refinement of the user requirements 30 can be promoted. By refining the user requirements 30, the proposed architecture 40 obtained from the large-scale generative model 2 can be refined. As a result, according to this embodiment, it can be expected that an appropriate architecture (proposed architecture 40) will be generated.

[0039] [Information Systems / Architecture] The information system may include any system related to information technology. The information system may include, for example, a cloud service, an on-premise system, a computer program, etc. The form of the information system may include any known form. For example, the cloud service may include AWS (Amazon Web Services), Cycloud, etc.

[0040] The architecture may include any components for achieving the purpose of the information system. The types of components of the information system are not particularly limited and may be appropriately selected depending on the embodiment. The components of the information system may include, for example, software, hardware, network, data, security, etc. The architecture may also include a basic structure between the components. The architecture may include, for example, a hardware configuration, a software configuration, a network configuration, a data structure, a security method, an interface design, etc. may include:

[0041] [User requirements] The user requirements 30 may be any requirements that user Z desires for an architecture related to an information system. The user requirements 30 may include the wants and demands of user Z. The user requirements 30 may be composed of dynamic information directly related to the architecture. The user requirements 30 may include, for example, one or more requirements such as "I want to create a matching app for connecting people based on hobbies," "I want to place a "Like" button on my profile page," "I want to be able to send messages," and "I want to be able to make restaurant reservations."

[0042] The user requirements 30 may be input in any data format, such as text, voice, image, etc. The voice data may be used as is, or may be converted into text by voice analysis and then used as the user requirements 30. At least a part of the user requirements 30 may be specified in natural language. The user requirements 30 are updated by answers 31 to the inquiries 49. Updating the user requirements 30 may include at least one of adding one or more new requirements and modifying one or more existing requirements. The new requirements may be requirements that are not included in the user requirements 30 before the update. The existing requirements may be requirements that are also included in the user requirements 30 before the update. The modification of the existing requirements may include at least one of deletion, modification, and replacement. In one example, the updated user requirements 30 (30B) may be configured to include the input answers 31 as is.

[0043] [First prompt] The first prompt P1 is a prompt given to the large-scale generative model 2 when generating the proposed architecture 40. The configuration of the first prompt P1 is not particularly limited and may be appropriately determined according to the embodiment as long as it includes the user requirements 30 and the proposed instructions 61. The first prompt P1 may further include any information other than the user requirements 30 and the proposed instructions 61.

[0044] As long as it is possible to instruct the large-scale generative model 2 to propose an architecture, the configuration of the proposal instruction 61 may not be particularly limited and may be appropriately determined depending on the embodiment. In one example, the proposal instruction 61 may include an instruction sentence instructing the proposal of an architecture that satisfies the given user requirement 30. The instruction sentence may be given in a natural language, for example, such as "Please propose an architecture that satisfies the user requirement." The data format of the instruction sentence of the proposal instruction 61 may not be particularly limited and may be appropriately selected depending on the embodiment. The generated proposed architecture 40 is an architecture proposed by the large-scale generative model 2.

[0045] [Second prompt] The second prompt P2 is a prompt to be given to the large-scale generative model 2 when generating a verification result 45 of the proposed architecture 40. The configuration of the second prompt P2 is not particularly limited and may be appropriately determined depending on the embodiment as long as it includes the proposed architecture 40 and the verification instruction 62. The second prompt P2 may further include any information other than the proposed architecture 40 and the verification instruction 62.

[0046] As long as it is possible to instruct the large-scale generative model 2 to verify the proposed architecture 40, the configuration of the verification instruction 62 is not particularly limited and may be appropriately determined depending on the embodiment. In one example, the verification instruction 62 may include an instruction statement instructing the verification of the proposed architecture 40. The instruction statement may be given in a natural language, such as "Please verify the proposed architecture". The data format of the instruction statement of the verification instruction 62 is not particularly limited and may be appropriately selected depending on the embodiment.

[0047] The generated verification result 45 is a result of verifying the proposed architecture 40 by the large-scale generative model 2. In one example, the verification result 45 may be output to the user Z. The verification result 45 may be output in a manner similar to the inquiry item 49, etc. Furthermore, the verification items (verification items) may be defined appropriately depending on the embodiment. At least a portion of the verification items may be specified in any manner. Alternatively, all of the verification items may be left to the large-scale generative model 2.

[0048] In one example, as illustrated in FIG. 3, the verification instruction 62 may include a verification list 620 including a plurality of predefined verification items. The verification items may be defined as appropriate depending on the embodiment. The verification items may include, for example, security, cost (operational cost, etc.), reliability (fault tolerance, etc.), scalability, performance, maintainability, manageability, etc. The maintainability may include CI (Continuous Integration), CD (Continuous Delivery / Continuous Deployment), etc. The manageability may include IAC (Infrastructure as Code), etc. For example, the verification instruction 62 may include the verification list 620 together with an instruction for instructing the verification of the proposed architecture 40.

[0049] Accordingly, the defects found in the verification may be, among the multiple verification items included in the verification list 620, verification items that are evaluated as having defects in the verification by the large-scale generative model 2. According to one example of the present embodiment, by defining the verification items in advance, it is possible to expect the scope of the verification by the large-scale generative model 2 to be guaranteed.

[0050] The deficiency may be any item that is or may be defective in the proposed architecture 40. The level of the deficiency may not be particularly limited and may be defined appropriately depending on the embodiment. The deficiency may include concerns, improvements, points to be noted, points to be noted, issues, pending items, etc. If one or more deficiencies are present in the verification result, generation of an inquiry item 49 by a third prompt P3 may be executed. At least a part of the one or more inquiry items 49 generated may relate to at least one of the one or more deficiencies. The one or more deficiencies included in the verification result 45 may include deficiencies related to items other than the verification items included in the verification list 620.

[0051] [Third prompt] The third prompt P3 is a prompt given to the large-scale generative model 2 when generating a query item 49 related to a deficiency. The configuration of the third prompt P3 is not particularly limited and may be appropriately determined depending on the embodiment as long as it includes the proposed architecture 40, the verification result 45, and the query generation instruction 63. The third prompt P3 may further include any information other than the proposed architecture 40, the verification result 45, and the query generation instruction 63.

[0052] As long as it is possible to instruct the large-scale generative model 2 to generate a query item 49 regarding defects in the proposed architecture 40, the configuration of the query generation instruction 63 need not be particularly limited and may be appropriately determined depending on the embodiment. In one example, the query generation instruction 63 may include an instruction sentence (first instruction sentence) to generate a query item 49 for user Z. The instruction sentence (first instruction sentence) may be given in a natural language, for example, "Please generate a query for the user". The data format of the instruction sentence (first instruction sentence) need not be particularly limited and may be appropriately selected depending on the embodiment.

[0053] The generated inquiry item 49 may include any item to inquire of the user Z about the deficiency item. The one or more generated inquiry items 49 may include an inquiry about an item other than the deficiency item. When the verification items are given as the verification list 620, the one or more generated inquiry items 49 may include an inquiry about an item other than the verification item. In one example, the generation of the query 49 may additionally employ at least one of the following four options:

[0054] (1) Creating inquiries based on skill level FIG. 6 is a schematic diagram showing an example of a method for generating an inquiry item 49 according to a skill level in this embodiment. In one example, the third prompt P3 may further include a skill level 351 of the user Z for designing an architecture. The skill level (skill level 351) may be an ability to understand terms related to the proposed architecture (proposed architecture 40). The method of expressing the skill level may not be particularly limited and may be appropriately determined according to the embodiment. The skill level may be given, for example, in ranks such as beginner, intermediate, and expert (advanced), or may be given in any other numerical indicator. The number of levels in the skill level may not be particularly limited and may be appropriately defined according to the embodiment. When expressing the skill level in ranks, names such as "beginner" may not be given. According to one example of this embodiment, it is expected that the inquiry item 49 will be generated according to the skill level 351 of the user Z. In addition, since the content of the inquiry item 49 is adjusted according to the skill level 351, it is expected that the possibility of obtaining an appropriate answer 31 will be improved.

[0055] (How to gain skill levels) The skill level 351 may be acquired in any manner. The skill level 351 may be given in advance by an answer from the user Z himself or another person, or may be obtained by inference in any manner. The skill level 351 of the user Z may be reflected in the input of the user Z. Therefore, the skill level 351 can be inferred from the input of the user Z. Thus, in one example, the skill level 351 may be inferred from the input 300 of the user Z.

[0056] The input 300 may include any response of the user Z. In one example, the input 300 may include at least a part of the user requirements 30. At a stage where the update of the proposed architecture 40 (re-proposition of the architecture) has been performed one or more times, the input 300 may include at least a part of the answer 31 to the inquiry item 49. The answer 31 may include a response 315 indicating that the answer is not possible (FIG. 7), which will be described later. The input 300 may include at least one of an answer to an evaluation of an alternative plan 451, which will be described later, an input specifying a part 318 that will not be changed, and an input specifying a part 319 that will be changed (FIGS. 9 and 10, which will be described later).

[0057] The input 300 may include responses other than the above responses (user requirements 30, answers 31, answers to the evaluation of the alternatives 451, inputs to specify the parts 318 not to be changed, inputs to specify the parts 319 to be changed, etc.) inputted when designing the architecture. For example, the architecture proposal device 1 may output a dedicated question to measure the skill level 351 of the user Z to the user Z, and accept an input of an answer to the dedicated question from the user Z. The content of the dedicated question may be appropriately determined according to the embodiment. The dedicated question may be outputted in a manner similar to the inquiry item 49. An input of the answer may be accepted in a manner similar to the user requirements 30, etc. The other responses may include this answer. The input 300 may be composed of at least one of the above responses and other responses inputted when designing the architecture.

[0058] According to one example of the present embodiment, by automatically inferring the skill level 351 from the input 300 of the user Z, it is expected that the effort required for the user Z to answer the skill level 351 can be reduced. Furthermore, in one example, since it does not depend on the subjective answer of the user Z, the objectivity of the obtained skill level 351 can be guaranteed. Therefore, it is expected that the third prompt P3 including this skill level 351 will improve the accuracy of generating the inquiry item 49 suitable for the user Z.

[0059] (Guessing method) The method of estimating the skill level 351 is not particularly limited and may be appropriately selected depending on the embodiment. The skill level 351 may be estimated by any method, such as language analysis, rule-based, or use of a trained model. The trained model may be appropriately generated by machine learning. As shown in FIG. 6, in one example, the architecture proposal device 1 may cause the large-scale generative model 2 to estimate the skill level 351 of the user Z by providing a fifth prompt P5 including an input 300 of the user Z and an instruction 65 to estimate the skill level 351.

[0060] The fifth prompt P5 is a prompt given to the large-scale generative model 2 when inferring the skill level 351 of user Z from the input 300 of user Z. The configuration of the fifth prompt P5 is not particularly limited and may be appropriately determined depending on the embodiment as long as it includes the input 300 and the inference instruction 65. The fifth prompt P5 may further include any information other than the input 300 and the inference instruction 65.

[0061] As long as it is possible to instruct the large-scale generative model 2 to estimate the skill level 351, the configuration of the inference instruction 65 may not be particularly limited and may be appropriately determined depending on the embodiment. In one example, the inference instruction 65 may include an instruction sentence instructing to estimate the skill level 351 from the input 300. The instruction sentence of the inference instruction 65 may be given in a natural language, such as "Please estimate the user's skill level from the user's input". The data format of the instruction sentence of the inference instruction 65 may not be particularly limited and may be appropriately selected depending on the embodiment. The skill level 351 obtained by inference (inference result) may be reflected in the third prompt P3 as appropriate.

[0062] The estimation of the skill level 351 by the large-scale generative model 2 may be performed at any time after any response is obtained from the user Z. In one example, the estimation of the skill level 351 may be performed at any time after the user requirements 30 are input and before the construction of the third prompt P3 is completed. In the case of assuming a stage after the update of the proposed architecture 40 is performed, the estimation of the skill level 351 may be performed at any time after the answer 31 to the inquiry item 49 is input and before the construction of the third prompt P3 is completed. In another example, when the answer to the dedicated question is used as the input 300, the estimation of the skill level 351 may be performed in advance before the input of the user requirements 30 is accepted.

[0063] According to one example of the present embodiment, by using the large-scale generative model 2 also to estimate the skill level 351, it is not necessary to prepare a dedicated model for estimating the skill level 351. Therefore, a reduction in costs can be expected. In addition, in one example, by using the large-scale generative model 2, the skill level 351 can be estimated while taking into account a wide range of contexts. Therefore, it can be expected that the estimation accuracy of the obtained skill level 351 is high.

[0064] At least a part of the process of estimating skill level 351 may be executed on the architecture proposal device 1, or may be executed on another computer other than the architecture proposal device 1. When adopting a form in which the whole process of estimating skill level 351 is executed by another computer, estimating skill level 351 may be configured by instructing the other computer to estimate skill level 351 and acquiring the estimation result of skill level 351 from the other computer.

[0065] (2) Correction of inquiry items It is not necessarily the case that user Z can properly answer all of the one or more inquiries 49 that are generated. The same is true for the above. Therefore, in one example, the architecture proposal device 1 may allow a response that the inquiry 49 (including the inquiry 49 generated according to the above skill level 351) is unanswerable. When the response that the inquiry 49 is unanswerable is made, the architecture proposal device 1 may cause the large-scale generative model 2 to modify the inquiry 49.

[0066] 7 is a schematic diagram showing an example of a method for correcting an inquiry item 49 in the present embodiment. In one example, the architecture proposal device 1 may accept input of an answer 31 to the inquiry item 49 in any format. In one example, the input format of the answer 31 may be determined according to the inquiry item 49, or may be predefined on the architecture proposal device 1. In either case, the input format of the answer 31 may be configured to allow input of a response 315 indicating that the question cannot be answered.

[0067] For example, when the input format of the answer 31 is determined according to the inquiry item 49, the inquiry generation instruction 63 may include an instruction sentence that allows the input of the response 315 that the answer cannot be answered. The instruction sentence that allows the response 315 that the answer cannot be answered may be incorporated into the first instruction sentence, or may be provided separately from the first instruction sentence. In this way, the inquiry item 49 that allows the response 315 that the answer cannot be answered may be generated. Also, for example, when the input format of the answer 31 is defined in advance, an input format that allows the response 315 that the answer cannot be answered may be prepared in advance. The input format may be defined on a program (program 81 in FIG. 21) or may be held as separate data in a memory resource. The architecture proposal device 1 may accept the input of the answer 31 in this input format, thereby allowing the input of the response 315 that the answer cannot be answered.

[0068] The response 315 indicating that an answer cannot be given may be input in any manner. In one example, the response 315 indicating that an answer cannot be given may be input as a free-form response such as "I don't know." In addition, the response 315 indicating that an answer cannot be given may be input as a selection-type response such as a checkbox. In addition, the unit for inputting the response 315 indicating that an answer cannot be given is not particularly limited, and may be appropriately determined depending on the embodiment. In one example, the response 315 indicating that an answer cannot be given may be input for each inquiry item 49, or may be input collectively for multiple inquiry items 49.

[0069] When the input to the inquiry 49 includes a response 315 indicating that the question cannot be answered, the architecture proposal device 1 may cause the large-scale generative model 2 to generate a modified inquiry 49 (49B) by providing a sixth prompt P6 including the inquiry 49 (49A), the skill level 351 of the user Z in architecture design, and a modification instruction 66. As described above, the skill level 351 may be acquired by any method. For convenience of explanation, the inquiry 49 before modification is also referred to as "inquiry 49A", and the inquiry 49 after modification is also referred to as "inquiry 49B".

[0070] The sixth prompt P6 is a prompt given to the large-scale generative model 2 when correcting the query 49 to match the skill level 351 of the user Z. The configuration of the sixth prompt P6 is not particularly limited and may be appropriately determined depending on the embodiment as long as it includes the query 49 (49A), the skill level 351, and the correction instruction 66. The sixth prompt P6 may further include any information other than the query 49 (49A), the skill level 351, and the correction instruction 66.

[0071] As long as it is possible to instruct the large-scale generative model 2 to modify the query item 49 according to the skill level 351, the configuration of the modification instruction 66 is not particularly limited and may be appropriately determined according to the embodiment. In one example, the modification instruction 66 may include an instruction statement for instructing to modify the query item 49. The instruction statement of the modification instruction 66 may be, for example, "Depending on the user's skill level, The instruction may be given in a natural language, such as "please correct the inquiry item by using the above information." The data format of the instruction text of the correction instruction 66 is not particularly limited, and may be appropriately selected depending on the embodiment.

[0072] The architecture proposal device 1 may output the modified inquiry item 49 (49B) to the user Z. The architecture proposal device 1 may accept input of a response 31 to the modified inquiry item 49 (49B) from the user Z. Updating the user requirements 30 may consist of updating the user requirements 30 with the response 31 input to the modified inquiry item 49 (49B). The output of the inquiry item 49 (49B) and the acceptance of the response 31 may be processed in the same manner as for the inquiry item 49 (49A) before modification. If a response 315 indicating that the modified inquiry item 49 (49B) is also unanswerable is input, modification of the inquiry item 49 may be repeated.

[0073] According to one example of the present embodiment, when an inquiry 49 that user Z cannot answer is generated, the inquiry 49 can be modified to match the skill level 351 of user Z, thereby making it possible to obtain a suitable answer 31 from user Z. In addition, in one example, a response 315 indicating that the inquiry cannot be answered is permitted, thereby preventing a perfunctory answer 31 based on vague knowledge from being input even when it is difficult to give a suitable answer 31. This is expected to increase the likelihood that the update of the proposed architecture 40 by the fourth prompt P4 using the input answer 31 will proceed in the appropriate direction.

[0074] In addition, when a response 315 indicating that an answer cannot be given to some of the pre-correction inquiry items 49 (49A) is input, and appropriate answers 31 are input to the rest, the input answers 31 may be handled arbitrarily. In one example, the input answers 31 may be reflected in updating the user requirements 30. That is, updating the user requirements 30 may include updating the user requirements 30 with the answers 31 input to the pre-correction inquiry items 49 (49A). In another example, the answers 31 input to the pre-correction inquiry items 49 (49A) may be discarded.

[0075] The method of correcting the inquiry item 49 may not be limited to such an example. The configuration of the sixth prompt P6 may not be limited to the above example, and may be appropriately changed depending on the embodiment. In another example, the sixth prompt P6 may further include a response 315 indicating that the question cannot be answered. In another example, the inquiry items 49 other than the inquiry item 49 to which the response 315 indicating that the question cannot be answered is input may be omitted from the sixth prompt P6. In another example, the sixth prompt P6 may be configured to instruct the user to correct the inquiry item 49 to a simpler expression without specifying the skill level 351. For example, the correction instruction 66 may include an instruction to simplify (lower the level of) the inquiry item 49. The instruction to simplify the inquiry item 49 may be incorporated into the instruction to correct the inquiry item 49, or may be provided separately from the instruction to correct the inquiry item 49. The wording of the instruction to simplify may be determined appropriately depending on the embodiment, for example, "Please make the expression easier to understand," "Please reduce the difficulty level of the terms used," etc. In this case, the skill level 351 may be omitted from the sixth prompt P6.

[0076] (3) Controlling the format of inquiries In one example, the format of the generated query item 49 may be appropriately controlled by including an arbitrary instruction in the query generation instruction 63. The instruction may be configured to instruct generation of the query item 49 in a target format. The content of the instruction may be appropriately determined depending on the control content. Three examples of controlling the format of the query item 49 are given below.

[0077] FIG. 8 illustrates a method for controlling the type of query 49 generated in this embodiment. An example is shown in schematic form. In one example, the query generation instruction 63 may include at least one of an instruction 631 (second instruction) for generating a query item 49 in which the input format of the answer 31 is restricted, an instruction 633 (third instruction) for listing the query items 49 in order of priority, and an instruction 635 (fourth instruction) for generating a query item in a format that asks for a request from user Z. Each instruction (631, 633, 635) may be incorporated into the first instruction or may be provided separately from the first instruction.

[0078] In one example, when the inquiry generation instruction 63 includes at least one of the instruction sentences (631, 633, 635), the above-mentioned form of generating the inquiry item 49 according to the skill level 351 may be adopted. That is, the third prompt P3 may further include the skill level 351. This skill level 351 may be omitted as appropriate. Also, in one example, when the inquiry generation instruction 63 includes at least one of the instruction sentences (631, 633, 635), the above-mentioned form of allowing input of a response 315 indicating that the generated inquiry item 49 cannot be answered and correcting the inquiry item 49 may be adopted. In this case, the correction instruction 66 for correcting the inquiry item 49 may also include at least one of the instruction sentences (631, 633, 635) like the generation instruction 63. In another example, the generation instructions 63 may include at least one of the directives (631, 633, 635), whereas the modification instructions 66 may omit the directives (631, 633, 635).

[0079] (Input format restrictions) When the generation instruction 63 includes the instruction 631, the restriction of the input format may be appropriately determined according to the embodiment. The restricted input format may include, for example, a selection format, a numerical input, etc. The selection format may include a single selection, multiple selection, range selection, date selection, etc. In the numerical input, a numerical range (upper limit, lower limit) may or may not be specified. The instruction 631 may be given in natural language, such as "Please generate an inquiry item in a selection format or a numerical input format." The data format of the instruction 631 may not be particularly limited and may be appropriately selected according to the embodiment. According to one example of this embodiment, the generation instruction 63 includes the instruction 631, so that an inquiry item 49 that is easy for the user Z to answer 31 can be obtained.

[0080] (Priority) When the generation instruction 63 includes the instruction 633, the priority may be set in any manner. In one example, the priority may be specified by the user Z, may be inferred from the input of the user Z (such as the input 300), may be set in advance in a program, or may be left to the large-scale generative model 2. The instruction 633 may be given in a natural language, such as "list them in order of priority." The data format of the instruction 633 is not particularly limited and may be appropriately selected according to the embodiment. According to one example of the present embodiment, the generation instruction 63 includes the instruction 633, so that the generated inquiry items 49 can be arranged in order of priority. This is expected to facilitate the answer 31 of the user Z.

[0081] Furthermore, when designing an architecture, it is possible to narrow the design scope and therefore to design the architecture more efficiently by starting with the parts that will have an impact on the whole (e.g., whether to use a server or go serverless) rather than starting with the details (e.g., how to take logs). In one example, the priority may be set according to the degree of impact on the design of this architecture. In this case, by arranging the inquiry items 49 in order of priority, it is possible to determine the answers 31 starting from the items with the greatest impact. This is expected to improve the efficiency of the design of the architecture.

[0082] (Request form) When the generation instruction 63 includes the instruction sentence 635, the request of the user Z includes a wish, a request, an intention, etc. The format of the request may be appropriately specified according to the embodiment. The format of the request may include, for example, question formats such as "Do you want to...?" and "Do you want to...?". The instruction sentence 635 may be given in a natural language, for example, "Please make the inquiry a question that confirms the user's request." The data format of the instruction sentence 635 is not particularly limited and may be appropriately selected according to the embodiment. According to the second experimental example described later, a result was obtained that by giving a prompt to instruct the generation of an inquiry item in the form of asking the user's request, it is highly likely that an appropriate inquiry item will be generated by the large-scale generative model. According to one example of the present embodiment, based on this knowledge, it can be expected that the generation instruction 63 includes the instruction sentence 635, thereby improving the possibility that an appropriate inquiry item 49 will be generated.

[0083] (others) The instruction included in the inquiry generation instruction 63 may not be limited to the above example, and may be appropriately changed depending on the embodiment. In another example, the inquiry generation instruction 63 may include an instruction to avoid a format asking about technical necessity, instead of each instruction (631, 633, 635) or together with at least one of the instruction (631, 633, 635). This instruction may be incorporated into the first instruction, or may be provided separately from the first instruction. The instruction may be provided in a natural language, such as "Please avoid questions asking about technical necessity". The data format of the instruction may not be particularly limited, and may be appropriately selected depending on the embodiment.

[0084] (4) Reflection of alternative proposals 9 is a schematic diagram showing an example of a scene in which an alternative 451 is generated in this embodiment. In one example, the architecture proposal device 1 may cause the large-scale generative model 2 to generate an alternative 451 for the proposed architecture 40 by providing a seventh prompt P7 including the proposed architecture 40, the verification result 45, and an instruction 67 for generating the alternative 451.

[0085] The seventh prompt P7 is a prompt to be given to the large-scale generative model 2 when generating an alternative 451 of the proposed architecture 40. The configuration of the seventh prompt P7 is not particularly limited and may be appropriately determined depending on the embodiment as long as it includes the proposed architecture 40, the verification result 45, and the generation instruction 67. The seventh prompt P7 may further include any information other than the proposed architecture 40, the verification result 45, and the generation instruction 67.

[0086] As long as it is possible to instruct the large-scale generative model 2 to generate an alternative 451 of the proposed architecture 40 in response to deficiencies found in the verification, the configuration of the generation instruction 67 is not particularly limited and may be appropriately determined depending on the embodiment. In one example, the generation instruction 67 may include an instruction statement instructing the generation of an alternative 451 of the proposed architecture 40. The instruction statement of the generation instruction 67 may be given in a natural language, such as "generate an alternative of the proposed architecture in consideration of the verification results." The data format of the instruction statement of the generation instruction 67 is not particularly limited and may be appropriately selected depending on the embodiment.

[0087] The obtained alternative 451 may be used arbitrarily. In one example, the obtained alternative 451 may be used to generate the query 49. That is, the third prompt P3 may further include the generated alternative 451. In this case, the generation of the alternative 451 may be performed at any time after the verification result 45 is obtained and before the construction of the third prompt P3 is completed. According to one example of the present embodiment, the generation of the alternative 451 can promote understanding of the proposed architecture 40. By including the generated alternative 451 in the third prompt P3, it is expected that the accuracy of the generation of the query 49 can be improved.

[0088] It should be noted that the method of using the alternative 451 is not limited to this example. In another example, The alternative 451 may not be used to generate the query 49 (construct the third prompt P3). In this case, the alternative 451 may be generated at any time after the verification result 45 is obtained.

[0089] Moreover, the architecture proposal device 1 may output the generated alternative 451 to the user Z, similarly to the inquiry item 49. Then, the architecture proposal device 1 may receive an input of an evaluation response to the alternative 451 from the user Z. The method of answering the evaluation may not be particularly limited and may be appropriately selected according to the embodiment. For example, the evaluation response may be expressed by a binary value of "good" or "bad". The number of evaluation ranks may not be limited to a binary value and may be appropriately changed according to the embodiment. Also, for example, the evaluation response may be obtained by a free-form description. In one example, the evaluation response to this alternative 451 may be interpreted as being included in the response 31 to the inquiry item 49. That is, the evaluation response to the alternative 451 may be reflected in the update of the user requirement 30. As a result, the fourth prompt P4 may further include an evaluation response to the alternative 451. According to one example of this embodiment, the large-scale generative model 2 can generate an updated proposed architecture 40 (re-propose the architecture) by further reflecting the evaluation response of the user Z to the alternative 451. Therefore, it is expected that the generation accuracy of the updated proposed architecture 40 can be improved.

[0090] In addition, in one example, when the form of using the alternative 451 to generate the inquiry item 49 is adopted, the above-mentioned form of generating the inquiry item 49 according to the above-mentioned skill level 351 may be further adopted. In this case, the third prompt P3 may further include the skill level 351 and the alternative 451. This skill level 351 may be omitted as appropriate. In one example, when the form of using the alternative 451 to generate the inquiry item 49 is adopted, the inquiry generation instruction 63 may include at least one of the above-mentioned instruction sentences (631, 633, 635). In one example, when the form of using the alternative 451 to generate the inquiry item 49 is adopted, the above-mentioned form of allowing input of the response 315 that the generated inquiry item 49 cannot be answered and correcting the inquiry item 49 may be adopted. In this case, the sixth prompt P6 when correcting the inquiry item 49 may also further include the alternative 451, similarly to the third prompt P3. In another example, the third prompt P3 further includes an alternative 451, whereas in the sixth prompt P6, the alternative 451 may be omitted.

[0091] [Fourth prompt] The fourth prompt P4 is a prompt given to the large-scale generative model 2 when updating the proposed architecture 40 (re-proposing the architecture). Updating the proposed architecture 40 may include modifying the proposed architecture 40. The configuration of the fourth prompt P4 is not particularly limited and may be appropriately determined according to the embodiment as long as it includes the updated user requirements 30 (30B), the proposed architecture 40 (40A), the verification results 45, the inquiry items 49, and the re-proposition instruction 64. The fourth prompt P4 may further include any information other than the updated user requirements 30 (30B), the proposed architecture 40 (40A), the verification results 45, the inquiry items 49, and the re-proposition instruction 64.

[0092] As long as it is possible to instruct the large-scale generative model 2 to update the proposed architecture 40 based on the answer 31 of the user Z to the inquiry item 49, the configuration of the re-proposal instruction 64 is not particularly limited and may be appropriately determined depending on the embodiment. In one example, the re-proposal instruction 64 may include an instruction sentence (first instruction sentence) that instructs re-proposing the architecture. The instruction sentence of the re-proposal instruction 64 may be given in a natural language, for example, "Please re-propose the architecture based on the user's answer". The data format of the instruction sentence of the re-proposal instruction 64 is not particularly limited and may be appropriately selected depending on the embodiment.

[0093] The generation process by this fourth prompt P4 updates the proposed architecture 40. Updating this proposed architecture 40 may include at least one of adding one or more new components and modifying one or more existing components. The new components may be components that are not included in the proposed architecture 40 (40A) before the update. The existing components may be components that are also included in the proposed architecture 40 (40A) before the update. The modification of the existing components may include at least one of deleting, changing, and replacing.

[0094] In this embodiment, the dialogue from verification to update of the proposed architecture 40 may be repeated any number of times until no deficiencies exist. This may lead to improvements in the proposed architecture 40. Note that no deficiencies exist may include not only that no deficiencies are found in the verification result 45, but also that the update of the proposed architecture 40 is discontinued at the discretion of the user Z.

[0095] (Update range) The range of updating the proposed architecture 40 is not particularly limited and may be appropriately determined depending on the embodiment. The range of updating the proposed architecture 40 may be left to the large-scale generative model 2 or may be determined by any information processing. In addition, the range of updating the proposed architecture 40 may be specified by the user Z.

[0096] 10 illustrates an example of a method for controlling the scope of updating the proposed architecture 40 (re-proposing the architecture) in this embodiment. In one example, the architecture proposal device 1 may output the generated proposed architecture 40 (40A) to the user Z. The proposed architecture 40 (40A) may be output in a manner similar to the inquiry item 49, etc.

[0097] The architecture proposal device 1 may receive an input from the user Z specifying at least one of the parts 318 to be left unchanged and the parts 319 to be changed in the proposed architecture 40 (40A). The input specifying at least one of the parts 318 to be left unchanged and the parts 319 to be changed may be received in a manner similar to that of the user requirements 30, etc. The parts 318 to be left unchanged and the parts 319 to be changed may be appropriately specified from among the components of the proposed architecture 40 (40A). The method of specifying each part (318, 319) is not particularly limited and may be appropriately determined according to the embodiment. The output of the proposed architecture 40 (40A) and the reception of the input specifying each part (318, 319) may be performed at any timing after the proposed architecture 40 (40A) is generated and before the construction of the fourth prompt P4 is completed.

[0098] When an input specifying the portion 318 to be left unchanged is received, the re-proposal instruction 64 may include an instruction 648 indicating that the specified portion 318 is not to be changed. When an input specifying the portion 319 to be changed is received, the re-proposal instruction 64 may include an instruction 649 indicating that the specified portion 319 is to be changed. When an input specifying both the portion 318 to be left unchanged and the portion 319 to be changed are received, the re-proposal instruction 64 may include instruction 648 and instruction 649.

[0099] As long as it is possible to instruct the large-scale generative model 2 not to change the specified portion 318, the configuration of the instruction statement 648 is not particularly limited and may be determined appropriately depending on the embodiment. The instruction statement 648 may be given in natural language, for example, "Please do not change the portion specified by the user." Also, as long as it is possible to instruct the large-scale generative model 2 to change the specified portion 319, the configuration of the instruction statement 649 is not particularly limited and may be determined appropriately depending on the embodiment. The instruction statement 649 may be given in natural language, for example, "Please do not change the portion specified by the user." Also, as long as it is possible to instruct the large-scale generative model 2 to change the specified portion 319, the configuration of the instruction statement 649 is not particularly limited and may be determined appropriately depending on the embodiment. The instruction sentences (648, 649) may be provided in a natural language such as "please change the part that has been re-proposed." Each instruction sentence (648, 649) may be incorporated into the first instruction sentence of the re-proposal instruction 64, or may be provided separately from the first instruction sentence. The data format of each instruction sentence (648, 649) is not particularly limited and may be appropriately selected depending on the embodiment.

[0100] In one example of this embodiment, when a designation of a part 318 that does not require change is received, the designated part 318 can be prevented from being changed when the architecture is re-proposed by reflecting the designation in the re-proposal instruction 64. Also, when a designation of a part 319 to be changed is received, the designation can be reflected in the re-proposal instruction 64, so that the designated part 319 can be prompted to be changed when the architecture is re-proposed.

[0101] Generate Summary 11 is a schematic diagram showing an example of a scene in which a summary 41 is generated in this embodiment. In one example, the architecture proposal device 1 may cause the large-scale generative model 2 to generate a summary 41 of the proposed architecture 40 by providing an eighth prompt P8 including the proposed architecture 40 and a summary instruction 68.

[0102] The eighth prompt P8 is a prompt to be given to the large-scale generative model 2 when generating a summary of the proposed architecture 40. The configuration of the eighth prompt P8 is not particularly limited and may be appropriately determined depending on the embodiment as long as it includes the proposed architecture 40 and the summarization instructions 68. The eighth prompt P8 may further include any information other than the proposed architecture 40 and the summarization instructions 68.

[0103] As long as it is possible to instruct the large-scale generative model 2 to generate a summary 41, the configuration of the summarization instruction 68 is not particularly limited and may be appropriately determined depending on the embodiment. In one example, the summarization instruction 68 may include an instruction sentence instructing the generation of a summary 41 of the proposed architecture 40. The instruction sentence of the summarization instruction 68 may be given in a natural language, for example, such as "Please generate a summary of the proposed architecture". The data format of the instruction sentence of the summarization instruction 68 is not particularly limited and may be appropriately selected depending on the embodiment. The generation of the summary 41 may be executed at any timing after the generation of the proposed architecture 40.

[0104] The generated summary 41 may be used in any manner. In one example, the architecture proposal device 1 may output the generated summary 41 to the user Z. The summary 41 may be output in a similar manner to the inquiry items 49, etc. The output of the summary 41 may be performed at any time after the generation of the summary 41. In one example, when a form of outputting the proposed architecture 40 is adopted, the summary 41 may be output together with the proposed architecture 40. Also, the summary 41 may be output separately from the proposed architecture 40, or may be output alone.

[0105] According to one example of the present embodiment, the output summary 41 can make it easier for user Z to understand the proposed architecture 40. By outputting the summary 41 together with the proposed architecture 40, the advantageous effect can be particularly expected. The configuration of the eighth prompt P8 may be appropriately changed depending on the embodiment. In another example, the eighth prompt P8 may further include a skill level 351. According to one example of the present embodiment, the summary 41 can be generated according to the skill level 351.

[0106] [Large-scale generative models] The large-scale generative model 2 may be a trained model generated by machine learning using a large amount of training data. The large-scale generative model 2 may include, for example, a large-scale language model, a large-scale visual language model, a large-scale voice model, etc. The large-scale generative model 2 may include a multi-modal model. The configuration and the number of parameters of the large-scale generative model 2 may vary depending on the embodiment. The large-scale generative model 2 may be, for example, a Transformer, a diffusion model, etc. The large-scale generative model 2 may have any structure. The large-scale generative model 2 may be fine-tuned depending on the purpose of use. As a specific example, the large-scale generative model 2 may be a known model such as Claude, GPT, or CyberAgentLM.

[0107] The large-scale generative model 2 may be deployed in any computer. In one example, the large-scale generative model 2 may be deployed in the architecture proposal device 1. In this case, the model data of the large-scale generative model 2 may be held in the memory resource of the architecture proposal device 1. The architecture proposal device 1 may execute each generation process by constructing each prompt P1 to P8 and providing each constructed prompt P1 to P8 to the large-scale generative model 2. In another example, the large-scale generative model 2 may be deployed in another computer (external server, etc.) other than the architecture proposal device 1. In this case, the architecture proposal device 1 may cause the large-scale generative model 2 to execute a generation process based on each prompt P1 to P8 by giving an instruction to the other computer. The architecture proposal device 1 causing the large-scale generative model 2 to generate each information (proposed architecture 40, etc.) may include the architecture proposal device 1 giving an instruction to the other computer to cause the other computer to execute a generation process of the large-scale generative model 2. In this case, at least a part of the process of constructing each prompt P1 to P8 may be executed on the architecture proposal device 1 or another computer. The architecture proposal device 1 may instruct the generation processes based on the prompts P1 to P8 individually, or may instruct at least two or more generation processes at once.

[0108] The data format of the information given as each of the prompts P1 to P8 is not particularly limited and may be appropriately selected depending on the embodiment. For example, each of the prompts P1 to P8 may be composed of data such as text, image, sound, token, etc. Tokens may be obtained by converting data such as text using a tokenizer. Each of the prompts P1 to P8 may be composed of multiple types of data. Each of the prompts P1 to P8 may further include any information other than the above components.

[0109] [Dialogue history] 12 is a schematic diagram showing an example of a scene in which a dialogue history 5 is generated in this embodiment. In one example, the architecture proposal device 1 may store the input 305 of the user Z, the instruction 600 for the large-scale generative model 2, and the generation result 400 of the large-scale generative model 2 in association with each other in chronological order as the dialogue history 5.

[0110] The input 305 of the user Z may include any response of the user Z. As long as the history of the responses of the user Z can be verified after the fact, the configuration of the input 305 is not particularly limited and may be appropriately determined according to the embodiment. In one example, the input 305 may include the user requirements 30 and the answers 31 to the inquiries 49. When adopting a form in which an input of an answer to a dedicated question is accepted in order to infer the skill level 351, the input 305 may include an answer to the dedicated question. When adopting a form in which an evaluation of the alternative 451 is accepted, the input 305 may include an answer to the evaluation of the alternative 451. When adopting a form in which an input of at least one of the part 318 not to be changed and the part 319 to be changed is accepted, the input 305 may include a designation (answer) of at least one of the part 318 not to be changed and the part 319 to be changed.

[0111] The instructions 600 may include any instructions for the large-scale generative model 2. As long as the history of the generation process of the large-scale generative model 2 can be verified after the fact, the configuration of the instructions 600 is not particularly limited and may be appropriately determined depending on the embodiment. In one example, the instructions 600 may include each of the prompts P1 to P8 as they are. The instructions 600 may also include each of the instructions 61 to 68. The instructions 600 may be used to simply execute the generation process according to each of the prompts P1 to P8 (each of the instructions 61 to 68). It may be configured as shown diagrammatically.

[0112] The generated result 400 may include any result of the generation process by the large-scale generative model 2. As long as the history of the result of the generation process by the large-scale generative model 2 can be verified after the fact, the configuration of the generated result 400 may not be particularly limited and may be appropriately determined according to the embodiment. In one example, the generated result 400 may include the proposed architecture 40, the verification result 45, and the inquiry item 49. When the inquiry item 49 is corrected, the generated result 400 may include the corrected inquiry item 49 (49B). When the large-scale generative model 2 is made to estimate the skill level 351, the generated result 400 may include the estimated skill level 351. When the large-scale generative model 2 is made to generate an alternative 451, the generated result 400 may include the alternative 451. When the large-scale generative model 2 is made to generate a summary 41, the generated result 400 may include the summary 41.

[0113] The storage area for storing the dialogue history 5 is not particularly limited and may be appropriately selected depending on the embodiment. The dialogue history 5 may be stored in at least one of a memory resource in the architecture proposal device 1, a storage medium, and an external storage device (including a memory resource of an external computer).

[0114] The timing for updating the dialogue history 5 (generating a history of each processing step) is not particularly limited and may be determined appropriately depending on the embodiment. In one example, the dialogue history 5 may be updated for each processing step. That is, each time a process such as an input by user Z or a generation process is executed, a corresponding history may be added to the dialogue history 5. Furthermore, the dialogue history 5 may be updated so as to add the histories of multiple processing steps all at once. The unit for generating the history may be determined appropriately depending on the embodiment.

[0115] According to one example of the present embodiment, by using the obtained dialogue history 5, it is possible to verify after the fact whether the generation process of the proposed architecture 40 is appropriate or not. Note that the use of the dialogue history 5 does not have to be limited to verification. The dialogue history 5 may be used in any situation other than verification. For example, at least a portion of the dialogue history 5 may be used to construct each of the prompts P1 to P8. In other words, each of the prompts P1 to P8 may include at least a portion of the dialogue history 5 obtained up to that point in time.

[0116] [Insider information] The method of managing various information such as the input of the user Z and the generation result of the large-scale generative model 2 is not particularly limited and may be appropriately determined depending on the embodiment. Various information may be internally stored as structured information. In one example, the architecture proposal device 1 may manage the user requirements 30, the proposed architecture 40, the verification results 45, and the inquiry items 49 as structured internal information.

[0117] When a form of obtaining user attributes including skill level 351 is adopted, the structured internal information may include the user attributes. When a form of generating alternatives 451 is adopted, the structured internal information may include alternatives 451. When a form of generating summary 41 is adopted, the structured internal information may include summary 41. When a form of generating dialogue history 5 is adopted, the structured internal information may include dialogue history 5.

[0118] 13 is a schematic diagram showing an example of the internal information 125 according to the present embodiment. In one example, the internal information 125 may include a user state 3, an architecture state 4, and a dialogue history 5.

[0119] The user state 3 at the target time may be composed of any information about the user Z that has been obtained up to the target time. In one example, the user state 3 is composed of the user requirements 30 (30A, 3 0B) and user attributes. In one example, the user requirements 30 may include answers 31 to inquiries 49. When a form for accepting an evaluation of an alternative 451 is adopted, the user requirements 30 may include an answer to the evaluation of the alternative 451. When a form for accepting an input for designating at least one of the part 318 not to be changed and the part 319 to be changed is adopted, the user requirements 30 may include a designation (answer) of at least one of the part 318 not to be changed and the part 319 to be changed.

[0120] The user attributes may be composed of any information of the user Z who may be involved in the design of the architecture or the interaction with the large-scale generative model 2. The user attributes may include the skill level 351 of the user Z. The user attributes may include any constraints by the user Z (e.g., hobbies, preferences, usage environment, etc.) other than the skill level 351. Each of the prompts P1 to P8 may include the user attributes. For example, the first prompt P1 includes the user attributes, so that the constraints of the user Z can be reflected in the generation of the proposed architecture 40. The fourth prompt P4 includes the user attributes, so that the constraints of the user Z can be reflected in the update of the proposed architecture 40. The second prompt P2 includes the user attributes, so that the constraints of the user Z can be reflected in the verification of the proposed architecture 40. In addition, the fourth prompt P4 includes the user attributes, so that the constraints of the user Z can be reflected in the generation of the inquiry item 49.

[0121] The architecture state 4 at the target time may be composed of any information about the architecture that has been obtained up to the target time. In one example, the architecture state 4 may include the proposed architecture 40 (40A, 40B), the verification result 45, and the inquiry items 49 (49A, 49B). When a form of generating a summary 41 is adopted, the architecture state 4 may include the generated summary 41. When a form of generating an alternative 451 is adopted, the architecture state 4 (verification result 45) may include the generated alternative 451. The dialogue history 5 at the target time may be composed of a history of dialogue between the user Z and the large-scale generative model 2 that has been obtained up to the target time. The dialogue history 5 may include an update history of the user state 3 and the architecture state 4.

[0122] The structuring format is not particularly limited and may be appropriately selected depending on the embodiment. In one example, a known format such as XML (extensible markup language) may be adopted as the structuring format. In addition, a storage area for holding the internal information 125 may be selected arbitrarily. The internal information 125 may be held in at least one of a memory resource in the architecture proposal device 1, a storage medium, and an external storage device (including a memory resource of an external computer). Each of the prompts P1 to P8 may include at least a part of the internal information 125 obtained up to that point in time, in addition to the above-mentioned components.

[0123] (Specific examples of internal information transfer) 14 to 20 show specific examples of transitions of the internal information 125 in this embodiment. In FIG. 14 to FIG. 20, an example of a scene is assumed in which a large-scale generative model 2 is made to estimate a skill level 351 before verification, a summary 41 is generated before verification, and an alternative 451 is generated. The internal information X in FIG. 14 to FIG. 20 is an example of the internal information 125. The user state A is an example of the user state 3. The architecture state B is an example of the architecture state 4. The dialogue history C is an example of the dialogue history 5. The code (0, 1, 2, etc.) after each piece of information (X, A, B, C) represents the transition.

[0124] (A) Initial state ~ Proposed architecture 14, in the initial state, the internal information X0 may be empty. That is, the user state A0, the architecture state B0, and the dialogue history C0 may each be empty. In response to receiving the input Z1 of the user requirement AA1, the architecture proposal device 1 may update the internal information X from X0 to X1. The user state A is the internal information X of the input user requirement A. The interaction history C may be updated from C0 to C1 to include the history that the user Z provided the input Z1 of the user requirement AA1 to the system (large-scale generative model 2) and the user state A was updated from A0 to A1.

[0125] A first prompt P1 may be constructed using the obtained user requirement AA1. A proposed architecture BA1 may be generated by providing the constructed first prompt P1 to the large-scale generative model 2. The generated proposed architecture BA1 may be output to the user Z. In response to the generation of the proposed architecture BA1, the architecture proposal device 1 may update the internal information X from X1 to X2. The architecture state B may be updated from B0 to B1 so as to include the generated proposed architecture BA1. The dialogue history C may be updated from C1 to C2 so as to include a history of the large-scale generative model 2 executing the generation process of the proposed architecture BA1, the architecture state B being updated from B0 to B1, and the proposed architecture BA1 being output.

[0126] (B) Inferring skill level / generating a summary As illustrated in FIG. 15, when the large-scale generative model 2 is to estimate the skill level, a fifth prompt P5 may be constructed using the obtained user input (input Z1) from the internal information X2. The constructed fifth prompt P5 may be provided to the large-scale generative model 2 to estimate the skill level ABA1. In response to the estimation of the skill level ABA1, the architecture proposal device 1 may update the internal information X from X2 to X2_1. The user state A may be updated from A1 to A1_1 so as to include the estimated skill level ABA1 (user attribute AB1). The dialogue history C may be updated from C2 to C2_1 so as to include a history of the large-scale generative model 2 executing the estimation process of the skill level ABA1 and the user state A being updated from A1 to A1_1.

[0127] Furthermore, when generating the summary 41, an eighth prompt P8 may be constructed using the obtained proposed architecture BA1 from the internal information X2 (X2_1). The constructed eighth prompt P8 may be given to the large-scale generative model 2 to generate a summary BB1. The generated summary BB1 may be output to the user Z. In response to the generation of the summary BB1, the architecture proposal device 1 may update the internal information X from X2 (X2_1) to X2_2. The architecture state B may be updated from B1 to B1_1 so as to include the generated summary BB1. The dialogue history C may be updated from C2 (C2_1) to C2_2 so as to include a history that the large-scale generative model 2 has executed the generation process of the summary BB1, that the architecture state B has been updated from B1 to B1_1, and that the summary BB1 has been output. Either the estimation of the skill level ABA1 or the generation of the summary BB1 may be executed first.

[0128] (C) Verification of the proposed architecture As illustrated in FIG. 16, a second prompt P2 may be constructed using the proposed architecture BA1 from the internal information X2 (X2_1, X2_2). A verification result BC1 of the proposed architecture BA1 may be generated by providing the constructed second prompt P2 to the large-scale generative model 2. The generated verification result BC1 may be output to the user Z. In response to the generation of the verification result BC1, the architecture proposal device 1 may update the internal information X from X2 (X2_1, X2_2) to X3. The architecture state B may be updated from B1 (B1_1) to B2 so as to include the generated verification result BC1. The dialogue history C may be updated from C2 (C2_1, C2_2) to C3 so as to include a history of the large-scale generative model 2 executing the generation process of the verification result BC1, the architecture state B being updated from B1 (B1_1) to B2, and the verification result BC1 being output.

[0129] (D) Generation of alternatives As illustrated in FIG. 17, a seventh prompt P7 may be constructed using the proposed architecture BA1 and the verification result BC1 from the internal information X3. An alternative BC1_1 of the proposed architecture BA1 may be generated by providing the constructed seventh prompt P7 to the large-scale generative model 2. The generated alternative BC1_1 may be output to the user Z. In response to the generation of the alternative BC1_1, the architecture proposal device 1 may update the internal information X from X3 to X3_1. The architecture state B may be updated from B2 to B2_1 to include the generated alternative BC1_1. The dialogue history C may be updated from C3 to C3_1 to include a history of the large-scale generative model 2 executing the generation process of the alternative BC1_1, the architecture state B being updated from B2 to B2_1, and the alternative BC1_1 being output. When an evaluation response to alternative BC1_1 is received, user state A (user requirements) may be updated to include the evaluation response to alternative BC1_1. Dialogue history C may be updated to include history that user Z has given an evaluation response and that user state A has been updated.

[0130] (E) Generation of inquiries As illustrated in FIG. 18, a third prompt P3 may be constructed using the proposed architecture BA1 and the verification result BC1 from the internal information X3 (X3_1). The third prompt P3 may be constructed to further include at least one of a user attribute AB1 (skill level ABA1) and an alternative BC1_1. A query BD1 may be generated by providing the constructed third prompt P3 to the large-scale generative model 2. The generated query 49 may be output to the user Z. In response to the generation of the query BD1, the architecture proposal device 1 may update the internal information X from X3 (X3_1) to X4. The architecture state B may be updated from B2 (B2_1) to B3 to include the generated query BD1. The dialogue history C may be updated from C3 (C3_1) to C4 to include history that the large-scale generative model 2 executed the generation process of query item BD1, that the architecture state B was updated from B2 (B2_1) to B3, and that query item BD1 was output.

[0131] (F) Receiving responses to inquiries As illustrated in FIG. 19, an input Z2 of a response 31 of a user Z to an outputted inquiry item BD1 may be obtained. In response to obtaining the input Z2 of the response 31, the architecture proposal device 1 may update the internal information X from X4 to X5. The user requirements AA may be updated from AA1 to AA2 so as to include the input Z2 of the response 31. In response to this, the user state A may be updated from A1 (A1_1) to A2. The dialogue history C may be updated from C4 to C5 so as to include a history of the user Z providing the input Z2 of the response 31 to the inquiry item BD1 to the system (large-scale generative model 2) and the user state A being updated from A1 (A1_1) to A2.

[0132] When an input specifying at least one of the parts 318 not to be changed and the parts 319 to be changed for the proposed architecture BA1 is received, the user state A (user requirements AA) may be updated to include a specification (answer) of at least one of the parts 318 not to be changed and the parts 319 to be changed. The dialogue history C may be updated to include a history that the user Z has given a specification of at least one of the parts 318 not to be changed and the parts 319 to be changed, and that the user state A has been updated.

[0133] Furthermore, if the input Z2 of the answer 31 includes a response 315 indicating that the answer is not possible, a sixth prompt P6 may be constructed using the query BD1. A modified query may be generated by providing the sixth prompt P6 to the large-scale generative model 2. The modified query may be output to the user Z. An input of an answer to the modified query may be accepted. In response to these, the architecture proposal device 1 Part information X may be updated. User state A (user requirements AA) may be updated to include a response to the modified query. Architecture state B may be updated to include the modified query. Dialogue history C may be updated to include history that large-scale generative model 2 performed generation processing of the modified query, that architecture state B was updated, that the modified query was output to user Z, that user Z provided input for the response to the modified query, and that the user state was updated.

[0134] (G) Re-proposition of architecture As illustrated in FIG. 20, a fourth prompt P4 may be constructed using the user requirements AA2, the proposed architecture BA1, the verification result BC1, and the inquiry item BD1 from the internal information X5. The constructed fourth prompt P4 may be given to the large-scale generative model 2 to generate an updated proposed architecture BA2. That is, the proposed architecture BA may be updated from BA1 to BA2. The updated proposed architecture BA2 may be output to the user Z. In response to obtaining the updated proposed architecture BA2, the architecture proposal device 1 may update the internal information X from X5 to X6. The architecture state B may be updated from B3 to B4 so as to include the updated proposed architecture BA2. The dialogue history C may be updated from C5 to C6 so as to include a history of the large-scale generative model 2 executing the generation process of the updated proposed architecture BA2, the architecture state B being updated from B3 to B4, and the updated proposed architecture BA2 being output. From this point on, the architecture proposal device 1 may appropriately update the internal information X until the re-proposition of the architecture is stopped.

[0135] According to one example of this embodiment, as shown in the above specific example, since the internal information 125 is structured, each prompt P1 to P8 can be easily constructed by extracting information of the target field. Therefore, it is expected that the implementation of the system (architecture proposing device) will be simplified. In addition, in one example, it is expected that the structured internal information 125 will simplify ex-post verification.

[0136] §2 Configuration Example [Hardware configuration] 21 is a schematic diagram showing an example of a hardware configuration of the architecture proposal device 1 according to the present embodiment. In one example, the architecture proposal device 1 may be configured as a computer to which a control unit 11, a storage unit 12, and a communication module 13 are electrically connected.

[0137] The control unit 11 may include a CPU, which is a hardware processor, a RAM (Random Access Memory), a ROM (Read Only Memory), etc., and is configured to execute information processing based on programs and various data. The control unit 11 (CPU) is an example of a processor resource.

[0138] The storage unit 12 may include, for example, a hard disk drive, a solid state drive, a semiconductor memory, etc., and is configured to hold any data. The storage unit 12, RAM, and ROM are examples of memory resources of the architecture proposal device 1. In one example of this embodiment, the storage unit 12 may store various information such as a program 81 and internal information 125. The program 81 is a program for causing the architecture proposal device 1 to execute information processing related to the architecture proposal (FIGS. 23 and 24 described below). The program 81 includes a series of instructions for the information processing. The program 81 is an example of a program (architecture proposal program) of the present disclosure.

[0139] In one example, at least one of the program 81 and the internal information 125 is stored in the storage unit 12. The program 81 may be stored in the storage medium 91 instead of or together with the storage unit 12. The storage medium 91 is configured to store various information (stored programs, etc.) by electrical, magnetic, optical, mechanical, or chemical action so that a machine such as a computer can read the information. The storage unit 12 and the storage medium 91 are examples of non-transient storage media. The architecture proposal device 1 may obtain the program 81 from the storage medium 91. The storage medium 91 may be a disk-type storage medium (CD, DVD, etc.) or a non-disk type storage medium such as a semiconductor memory (flash memory, etc.). An arbitrary drive device may be used to read the information stored in the storage medium 91. The type of the drive device may be selected according to the storage medium 91. The drive device may be connected to the architecture proposal device 1 in an arbitrary manner. The storage medium 91 may include an external storage device.

[0140] The communication module 13 is configured to perform wired or wireless communication via a network. The communication module 13 may be configured, for example, by a wired LAN (Local Area Network) module, a wireless LAN module, or the like. The network standard is not particularly limited and may be appropriately selected depending on the embodiment. For example, the type of network may be appropriately selected from the Internet, a wireless communication network, a mobile communication network, a telephone network, a dedicated network, or the like. The architecture proposal device 1 may perform data communication with another computer (for example, a user terminal U1, an external computer, or the like) via the communication module 13.

[0141] In addition, regarding the specific hardware configuration of the architecture proposal device 1, components can be omitted, replaced, and added as appropriate depending on the embodiment. For example, the control unit 11 may include a plurality of hardware processors. The hardware processor may be configured with a microprocessor, a field-programmable gate array (FPGA), a digital signal processor (DSP), a graphics processing unit (GPU), an application specific integrated circuit (ASIC), or the like. The architecture proposal device 1 may further include an input device and an output device. The input device is configured to receive input of information. The input device may be configured with, for example, a mouse, a keyboard, an operator, or the like. The output device is configured to output information. The output device may be configured with, for example, a display, a speaker, or the like. The input device and the output device may be integrally configured at least in part with a touch panel display or the like. The input device and the output device may be appropriately connected to the architecture proposal device 1. The communication module 13 may be omitted. At least one of the program 81 and the internal information 125 may be stored in an external storage device such as a network attached storage (NAS). An external storage device is also an example of a non-transitory storage medium. The architecture proposal device 1 may be configured with multiple computers. In this case, the hardware configurations of the computers may or may not be the same. The architecture proposal device 1 may be a computer designed specifically for the service to be provided, as well as a general-purpose server device, a general-purpose PC, a notebook PC, a terminal device, etc. The terminal device may include a smartphone, a tablet terminal, etc.

[0142] [Software configuration] 22 is a schematic diagram showing an example of a software configuration of the architecture proposal device 1 according to the present embodiment. The control unit 11 of the architecture proposal device 1 executes instructions included in the program 81 stored in the storage unit 12 by the CPU. As a result, the architecture proposal device 1 operates as a computer including a requirement receiving unit 1101, a first request unit 1102, a second request unit 1103, a third request unit 1104, an output processing unit 1105, a response receiving unit 1106, a designation receiving unit 1107, a fourth request unit 1108, a level estimation unit 1109, a sixth request unit 1110, a seventh request unit 1111, an eighth request unit 1112, and a history generation unit 1113 as software modules. That is, in one example of the present embodiment, each software module of the architecture proposal device 1 is realized by the control unit 11 (CPU).

[0143] The requirement receiving unit 1101 is configured to receive an input of user requirements 30 for an architecture related to an information system from a user Z. The first request unit 1102 is configured to cause the large-scale generative model 2 to generate a proposed architecture 40 of the information system by providing a first prompt P1 including the input user requirements 30 and a proposal instruction 61. The second request unit 1103 is configured to cause the large-scale generative model 2 to generate a verification result 45 of the proposed architecture 40 by providing a second prompt P2 including the generated proposed architecture 40 and a verification instruction 62. The third request unit 1104 is configured to cause the large-scale generative model 2 to generate a query item 49 by providing a third prompt P3 including the proposed architecture 40, the verification result 45, and a query generation instruction 63 when a deficiency exists.

[0144] The output processing unit 1105 is configured to output the generated inquiry item 49 to the user Z. The answer receiving unit 1106 is configured to receive an input of an answer 31 to the inquiry item 49 from the user Z, and to update the user requirements 30 with the input answer 31. The fourth request unit 1108 is configured to cause the large-scale generative model 2 to generate an updated proposed architecture 40 by providing a fourth prompt P4 including the updated user requirements 30, the proposed architecture 40, the verification result 45, the inquiry item 49, and a re-proposal instruction 64.

[0145] Until no deficiencies exist, the generation of the verification result 45 by the second request unit 1103, the generation of the inquiry item 49 by the third request unit 1104, the output of the inquiry item 49 by the output processing unit 1105, the acceptance of the input of the answer 31 by the answer acceptance unit 1106, the update of the user requirement 30 by the answer acceptance unit 1106, and the update of the proposed architecture 40 by the fourth request unit 1108 may be repeated. This may improve the proposed architecture 40. The output processing unit 1105 is configured to output the improved proposed architecture 40.

[0146] The output processing unit 1105 may be further configured to output the generated proposed architecture 40 to the user Z. The designation receiving unit 1107 is configured to receive an input from the user Z designating at least one of a portion 318 to be left unchanged and a portion 319 to be changed in the proposed architecture 40. The fourth request unit 1108 may be configured to cause the large-scale generative model 2 to generate an updated proposed architecture 40 using a re-proposal instruction 64 including at least one of an instruction 648 indicating that the designated portion 318 is not to be changed and an instruction 649 indicating that the designated portion 319 is to be changed.

[0147] The level estimation unit 1109 is configured to estimate the skill level 351 of user Z from the input 300 of user Z. The level estimation unit 1109 may include a fifth request unit 11091. The fifth request unit 11091 is configured to cause the large-scale generative model 2 to estimate the skill level 351 of user Z by providing a fifth prompt P5 including the input 300 of user Z and an instruction 65 for estimating the skill level 351.

[0148] The sixth request unit 1110 is configured to, when the input to the inquiry item 49 includes a response 315 indicating that the inquiry item 49 is unanswerable, cause the large-scale generative model 2 to generate a modified inquiry item 49 by providing a sixth prompt P6 including the inquiry item 49, a skill level 351, and a modification instruction 66. The output processing unit 1105 may be configured to output the modified inquiry item 49 to the user Z. The answer receiving unit 1106 may be configured to receive an input of an answer 31 to the modified inquiry item 49 from the user Z, and to update the user requirements 30 with the input answer 31.

[0149] The seventh request unit 1111 issues a seventh prompt P7 including the proposed architecture 40, the verification result 45, and an instruction 67 for generating an alternative 451, thereby generating an alternative 451 for the proposed architecture 40. The third request unit 1104 may be configured to cause the large-scale generative model 2 to generate a query item 49 by using a third prompt P3 that further includes the generated alternative 451.

[0150] The eighth request unit 1112 is configured to cause the large-scale generative model 2 to generate a summary 41 of the proposed architecture 40 by providing an eighth prompt P8 including the proposed architecture 40 and a summary instruction 68. The output processing unit 1105 may be further configured to output the generated summary 41 to the user Z. The history generation unit 1113 is configured to store the input 305 of the user Z, the instruction 600 for the large-scale generative model 2, and the generation result 400 of the large-scale generative model 2 in association with each other in chronological order as a dialogue history 5.

[0151] §3 Example of operation 23 and 24 are flowcharts showing an example of a processing procedure for proposing an architecture by the architecture proposal device 1 according to this embodiment. The following processing procedure is an example of an information processing method (architecture proposal method) of the present disclosure. However, the following processing procedure is merely an example, and each step may be modified as much as possible. Furthermore, steps in the following processing procedure may be omitted, replaced, or added as appropriate depending on the embodiment.

[0152] (Step S101) In step S101, the control unit 11 operates as a requirement receiving unit 1101 and receives an input of user requirements 30 for an architecture related to an information system from a user Z. In one example, the control unit 11 may receive the input of the user requirements 30 via a user terminal U1 connected to the architecture proposal device 1. Upon receiving the input of the user requirements 30, the control unit 11 advances the process to the next step S102.

[0153] (Step S102) In step S102, the control unit 11 operates as a first request unit 1102 and provides a first prompt P1 including the input user requirements 30 and the proposed instruction 61 to the large-scale generative model 2. As a result, the control unit 11 causes the large-scale generative model 2 to generate the proposed architecture 40. After generating the proposed architecture 40, the control unit 11 proceeds to the next step S103.

[0154] (Step S103) In step S103, the control unit 11 operates as a level estimation unit 1109 and estimates the skill level 351 of the user Z from the input 300 of the user Z. The method of estimating the skill level 351 may be appropriately selected according to the embodiment. In one example, the control unit 11 operates as a fifth request unit 11091 and may give the large-scale generative model 2 a fifth prompt P5 including the input 300 of the user Z and an instruction 65 to estimate the skill level 351. In this way, the control unit 11 may cause the large-scale generative model 2 to estimate the skill level 351 of the user Z. Note that the timing of executing the process of step S103 is not limited to such an example. The process of step S103 may be executed at any timing. In addition, the control unit 11 may execute the process of step S103 after the input 300 of the user Z is accumulated beyond a predetermined amount. After estimating the skill level 351, the control unit 11 advances the process to the next step S104.

[0155] (Step S104) In step S104, the control unit 11 operates as an eighth request unit 1112 and provides an eighth prompt P8 including the proposed architecture 40 and a summary instruction 68 to the large-scale generative model 2. As a result, the control unit 11 causes the large-scale generative model 2 to generate a summary 41 of the proposed architecture 40. Note that the timing for executing the process of step S104 may vary depending on the example. The process of step S104 may be executed at any timing after the process of step S102 has been executed. After generating the summary 41, the control unit 11 advances the process to the next step S105.

[0156] (Step S105) In step S105, the control unit 11 operates as a second request unit 1103 and provides a second prompt P2 including the generated proposed architecture 40 and a verification instruction 62 to the large-scale generative model 2. As a result, the control unit 11 causes the large-scale generative model 2 to generate a verification result 45 of the proposed architecture 40.

[0157] In one example, the verification instruction 62 may include a verification list 620 including a plurality of predefined verification items. In this way, when a deficiency is found in the verification, at least a portion of the found deficiency may be a verification item that is evaluated as having a deficiency in the verification by the large-scale generative model 2, among the plurality of verification items included in the verification list 620. In another example, the control unit 11 may operate as an output processing unit 1105 at any timing after executing the process of step S105, and output the generated verification result 45 to the user Z. When the verification result 45 is generated, the control unit 11 proceeds to the next step S106.

[0158] (Step S106) In step S106, the control unit 11 operates as an output processing unit 1105 and outputs the generated proposed architecture 40 and abstract 41 to user Z. Note that the timing of outputting the proposed architecture 40 and abstract 41 is not limited to this example. The proposed architecture 40 may be output at any timing after the processing of step S102 is executed. The abstract 41 may be output at any timing after the processing of step S104 is executed. The proposed architecture 40 and the abstract 41 may be output separately. After outputting the proposed architecture 40 and the abstract 41, the control unit 11 proceeds to the next step S107.

[0159] (Step S107) In step S107, the control unit 11 determines the branch destination of the process depending on whether or not one or more deficiencies exist in the verification result 45. If one or more deficiencies exist, the control unit 11 advances the process to the next step S108. On the other hand, if no deficiencies exist, the control unit 11 ends the process procedure related to this operation example.

[0160] In one example, the control unit 11 may determine whether or not one or more defects exist according to the generated verification result 45. The control unit 11 may determine a branch destination of the process according to the determination result. In another example, the control unit 11 may receive an operation from the user Z (user terminal U1) as to whether or not one or more defects exist. The control unit 11 may determine a branch destination of the process according to the result of the operation of the user Z. In response to the user Z performing an operation to request an update of the proposed architecture 40, the control unit 11 may determine that one or more defects exist, and proceed to the next step S108. On the other hand, in response to the user Z performing an operation to abort the update of the proposed architecture 40, the control unit 11 may determine that no defects exist, and end the processing procedure related to this operation example. The operation to request an update and the operation to abort the update are not particularly limited, and may be appropriately determined according to the embodiment.

[0161] (Step S108) In step S108, the control unit 11 operates as a seventh request unit 1111 and provides a seventh prompt P7 including the proposed architecture 40, the verification result 45, and an instruction 67 for generating an alternative 451 to the large-scale generative model 2. As a result, the control unit 11 causes the large-scale generative model 2 to generate an alternative 451 for the proposed architecture 40.

[0162] In one example, the control unit 11 may operate as an output processing unit 1105 at any timing after executing the process of step S108, and output the generated alternative 451 to the user Z. The control unit 11 may operate as a response receiving unit 1106, and receive an input of an evaluation response to the alternative 451 from the user Z, and update the user requirement 30 with the input evaluation response. Note that the timing of executing the process of step S108 is not limited to this example. The process of step S108 may be executed at any timing after executing the process of step S105. After generating the alternative 451, the control unit 11 advances the process to the next step S109.

[0163] (Step S109) In step S109, the control unit 11 operates as a third request unit 1104 and provides a third prompt P3 including the proposed architecture 40, the verification result 45, and a query generation instruction 63 to the large-scale generative model 2. As a result, the control unit 11 causes the large-scale generative model 2 to generate a query item 49.

[0164] In one example, the constructed third prompt P3 may further include a skill level 351 of user Z. In one example, the third prompt P3 may further include an alternative 451. In another example, the query generation instruction 63 may include at least one of an instruction 631 for generating a query item 49 in which the input format of the answer 31 is restricted, an instruction 633 for listing the query items 49 in order of priority, and an instruction 635 for generating a query item in a format that asks for the request of user Z. After generating the query item 49, the control unit 11 proceeds to the next step S110.

[0165] (Step S110-Step S111) In step S110, the control unit 11 operates as the output processing unit 1105 and outputs the generated inquiry item 49 to the user Z. In one example, the control unit 11 may output the inquiry item 49 to the user terminal U1. After outputting the inquiry item 49, the control unit 11 advances the process to the next step S111.

[0166] In step S111, the control unit 11 operates as the answer receiving unit 1106 and receives input of the answer 31 to the inquiry item 49 from the user Z. In one example, the control unit 11 may receive the input of the answer 31 via the user terminal U1. Upon receiving the input of the answer 31, the control unit 11 advances the process to the next step S112.

[0167] (Step S112) In step S112, the control unit 11 determines where the process should branch depending on whether the input answer 31 to the inquiry item 49 includes a response 315 indicating that the answer cannot be answered. If the answer 31 includes the response 315 indicating that the answer cannot be answered, the control unit 11 proceeds to step S113. On the other hand, if the answer 31 does not include the response 315 indicating that the answer cannot be answered, the control unit 11 proceeds to step S114.

[0168] (Step S113) In step S113, the control unit 11 operates as a sixth request unit 1110 and provides a sixth prompt P6 including the query item 49, the skill level 351, and the modification instruction 66 to the large-scale generative model 2. As a result, the control unit 11 causes the large-scale generative model 2 to generate the modified query item 49. After generating the modified query item 49, the control unit 11 returns the process to step S110.

[0169] In step S110, the control unit 11 operates as the output processing unit 1105 and outputs the corrected The inquiry item 49 is output to the user Z. In step S111, the control unit 11 accepts input of a reply 31 to the revised inquiry item 49 from the user Z. The control unit 11 may repeatedly execute the processes of steps S113, S110 and S111 until there are no more responses 315 indicating that the reply is not possible.

[0170] (Step S114) In step S114, the control unit 11 operates as the answer receiving unit 1106, and updates the user requirements 30 with the input answer 31. When the process of step S113 is being executed, the control unit 11 may update the user requirements 30 with the answer 31 to the corrected inquiry item 49. After updating the user requirements 30, the control unit 11 proceeds to the next step S115.

[0171] (Step S115) In step S115, the control unit 11 operates as the designation receiving unit 1107 and receives an input from the user Z designating at least one of the part 318 to be left unchanged and the part 319 to be changed in the proposed architecture 40. In one example, the control unit 11 may receive the designation of at least one of the part 318 to be left unchanged and the part 319 to be changed via the user terminal U1. The control unit 11 updates the user requirement 30 according to the designation (answer) of at least one of the part 318 to be left unchanged and the part 319 to be changed in the proposed architecture 40. Note that the timing of executing the processes of steps S106 and S115 is not limited to this example. The process of step S115 may be executed at any timing after the proposed architecture 40 is output. The process of steps S106 and S115 may be executed before the process of steps S110 and S111. The process of steps S106 and S115 may be executed at least partially in parallel with the process of steps S110 and S111. After updating the user requirement 30, the control unit 11 advances the process to the next step S116.

[0172] (Step S116) In step S116, the control unit 11 operates as a fourth request unit 1108 and provides a fourth prompt P4 including the updated user requirements 30, the proposed architecture 40, the verification result 45, the inquiry item 49, and the re-proposal instruction 64 to the large-scale generative model 2. As a result, the control unit 11 causes the large-scale generative model 2 to generate the updated proposed architecture 40. In one example, the re-proposal instruction 64 may include at least one of an instruction statement 648 indicating that the specified part 318 is not to be changed and an instruction statement 649 indicating that the specified part 319 is to be changed. When the updated proposed architecture 40 is generated, the control unit 11 returns the process to step S103 and executes the process again from step S103.

[0173] The control unit 11 may repeatedly execute the processes from step S103 to step S116 until it is determined in the process of step S107 that there is no deficiency. In this way, the control unit 11 may improve the proposed architecture 40. The control unit 11 may output the improved proposed architecture 40 by the process of step S106. The control unit 11 may operate as a history generation unit 1113 at any timing, and may associate the input 305 of the user Z, the instruction 600 for the large-scale generative model 2, and the generation result 400 of the large-scale generative model 2 in chronological order as the dialogue history 5 and store them. The control unit 11 may generate a history for each step, or may generate a history of multiple steps at once. In addition, the control unit 11 may manage various information obtained in each step (user requirements 30, proposed architecture 40, verification results 45, inquiry items 49, etc.) as structured internal information 125.

[0174] [Features] In this embodiment, the architecture proposal device 1 is a device for proposing an architecture between a user Z and a large-scale generative model 2. As an alternating dialogue, architecture proposal (steps S102, S116), verification (step S105), and inquiries to user Z (steps S101, S111) are repeatedly executed. In one example, the verification and inquiries are automatically and repeatedly executed. In this dialogue, the answers 31 to the inquiries 49 are repeated, which can promote the refinement of the user requirements 30. By refining the user requirements 30, the proposed architecture 40 obtained in the processing of step S116 can be refined. As a result, according to this embodiment, it can be expected that an appropriate architecture (proposed architecture 40) will be generated.

[0175] § 4 Variations Although the embodiment of the present disclosure has been described in detail above, the above description is merely an example of the present disclosure in every respect. The processes and means described in the present disclosure can be freely combined and implemented as long as no technical contradiction occurs. In addition, various improvements or modifications may be made to the above embodiment as appropriate. For example, the following modifications are possible. In the following, the same reference numerals are used for components similar to those in the above embodiment, and the description of the same points as those in the above embodiment is omitted as appropriate.

[0176] At least one of omission, replacement, and addition of steps may be performed on the processing procedure of the architecture proposal device 1 according to the above embodiment. For example, the process of generating the dialogue history 5 may be omitted. In response to this, the history generation unit 1113 may be omitted from the software configuration of the architecture proposal device 1. Also, for example, the process of step S104 and the output of the summary 41 may be omitted. In response to this, the eighth request unit 1112 may be omitted from the software configuration of the architecture proposal device 1. Also, for example, the process of step S108 may be omitted. In response to this, the seventh request unit 1111 may be omitted from the software configuration of the architecture proposal device 1. Also, for example, the processes of steps S112 and S113 may be omitted. In response to this, the sixth request unit 1110 may be omitted from the software configuration of the architecture proposal device 1. Also, for example, the process of step S106 may be omitted. Also, for example, the process of step S103 may be omitted. The skill level 351 may be given in advance. The skill level 351 may be omitted. Accordingly, the level estimation unit 1109 (the fifth request unit 11091) may be omitted from the software configuration of the architecture proposal device 1. Also, for example, the processing of step S115 may be omitted. Accordingly, the designation receiving unit 1107 may be omitted from the software configuration of the architecture proposal device 1.

[0177] §5 Experimental Examples The following experiments were carried out to verify the effectiveness of the above embodiment, however, the present invention is not limited to the following experimental examples.

[0178] [First Experimental Example] First, as a first experimental example, it was verified that an appropriate architecture can be obtained by repeating the process of proposing, verifying, querying, and re-proposing an architecture using a large-scale generative model in a manner similar to that of the above embodiment.

[0179] Specifically, we used GPT-4o (gpt-4o-2024-08-06) for the large-scale generative model. The requirements given were, "I want to create a matching app using AWS to connect people with similar hobbies. I want to be able to quickly and easily look at profiles and like them, and send messages when a match occurs. I also want to be able to make restaurant reservations." Using these user requirements, we constructed the first prompt. By feeding the constructed first prompt to a large-scale generative model, we generated the proposed architecture (we obtained the first proposed architecture).

[0180] The second prompt was constructed using the generated proposed architecture. The constructed second prompt was fed into a large-scale generative model to generate verification results. The third prompt was constructed using the obtained proposed architecture and verification results. The constructed third prompt was fed into a large-scale generative model to generate inquiries. Answers to the generated inquiries were input. The fourth prompt was constructed using the user requirements, proposed architecture, verification results, and inquiries updated with the input answers. The constructed fourth prompt was fed into a large-scale generative model to update the proposed architecture. In other words, the second proposed architecture (the result of the first re-proposal of the architecture) was obtained.

[0181] The second prompt was constructed again using the updated proposed architecture. The constructed second prompt was fed into the large-scale generative model to generate validation results again. The third prompt was constructed again using the updated proposed architecture and validation results. The constructed third prompt was fed into the large-scale generative model to generate queries again. Answers to the generated queries were input. The fourth prompt was constructed again using the user requirements, proposed architecture, validation results, and queries updated with the input answers. The constructed fourth prompt was fed into the large-scale generative model to update the proposed architecture. In other words, the third proposed architecture (the result of the second re-proposition of the architecture) was obtained.

[0182] 25 and 26 show the architecture proposed for the first time in the first experimental example. FIG. 27 shows the verification result of the architecture proposed for the first time in the first experimental example. FIG. 28 shows the inquiry and the answer regarding the architecture proposed for the first time in the first experimental example. FIG. 29 and 30 show the architecture proposed for the second time in the first experimental example (the result of the first re-proposal of the architecture). FIG. 31 shows the verification result of the architecture proposed for the second time in the first experimental example. FIG. 32 shows the inquiry and the answer regarding the architecture proposed for the second time in the first experimental example. FIG. 33 and 34 show the architecture proposed for the third time in the first experimental example (the result of the second re-proposal of the architecture). Comparing the first proposed architecture (FIG. 25 and FIG. 26) and the third proposed architecture (FIG. 33 and FIG. 34), it was possible to obtain an appropriate architecture by refining the user requirements. From this result, it was verified that it is possible to expect the generation of an appropriate architecture according to this embodiment.

[0183] [Second Experimental Example] Next, we verified that appropriate queries could be obtained by including in the generation instructions an instruction to generate queries in the form of a user request. As in the first experimental example, we used GPT-4o (gpt-4o-2024-08-06) for the large-scale generative model. As in the first experimental example, we used the initial user By providing user requirements, the process proceeded to the generation of a query. When generating this query, in experimental example 2-1, a prompt that did not include an instruction to generate a query in the form of a user request was provided to the large-scale generative model. On the other hand, in experimental example 2-2, when generating a query, a prompt that included an instruction to generate a query in the form of a user request was provided to the large-scale generative model.

[0184] Figure 35 shows the instruction given to the large-scale generative model as an instruction to generate a query in the form of a user request. Figure 36 shows the result of query generation obtained by giving a prompt that does not include an instruction to generate a query in the form of a user request (Experimental Example 2-1). Figure 37 shows the result of query generation obtained by giving a prompt that includes an instruction to generate a query in the form of a user request (Experimental Example 2-2). When the prompt does not include an instruction to generate a query in the form of a user request, the query "Is the availability of cloud services important?" is generated. In contrast, when the prompt included an instruction to generate a query in the form of a user request, the query generated was "Do you want to improve the availability of the application?". In cases where the prompt included an instruction to generate a query in the form of a user request, including other queries, the resulting query was more appropriate than when the prompt did not include the instruction. This result shows that by including an instruction to generate a query in the form of a user request in the generation instructions, it is possible to increase the likelihood that an appropriate query will be generated. [Explanation of symbols]

[0185] 1. Architecture proposal device, 2. Large-scale generative model, 30...User requirements, 61...Proposal instructions, 40...Proposed architecture, 62...Verification instructions, 45... Verification result, 63... Generation instruction, 49…Inquiries, 31...Answer, 64...Reproposal instruction, P1: 1st prompt, P2: 2nd prompt, P3…3rd prompt, P4…4th prompt, Z: User

Claims

1. accepting input of user requirements for an architecture relating to the information system from a user; generating a proposed architecture for the information system in a large-scale generative model upon providing a first prompt including the input user requirements and proposed instructions; causing the large-scale generative model to generate a validation result for the proposed architecture by providing a second prompt including the generated proposed architecture and a validation instruction; if there is a deficiency in the generated verification result, generating a query regarding the deficiency in the large-scale generative model by providing a third prompt including the proposed architecture, the verification result, and an instruction to generate a query; outputting the generated query to the user; accepting input of a response to the inquiry from the user; updating the user requirements with the input answers; causing the large-scale generative model to generate an updated proposed architecture by providing a fourth prompt including the updated user requirements, the proposed architecture, the validation results, the query, and a re-proposal instruction; refining the proposed architecture by repeating the steps of generating verification results for the proposed architecture, generating queries, outputting the queries, accepting input of responses to the queries, updating the user requirements, and generating the updated proposed architecture until no deficiencies exist; and outputting an improved version of the proposed architecture; A control unit configured to execute Architecture proposal device.

2. the verification instructions include a verification list including a plurality of predefined verification items; The deficiency item is a verification item that is evaluated as having a deficiency in verification by the large-scale generative model among a plurality of verification items included in the verification list. The architecture proposal apparatus according to claim 1 .

3. the third prompt further includes a skill level of the user with respect to designing the architecture. The architecture proposal apparatus according to claim 1 .

4. The user's skill level is inferred from the user's input. The architecture proposal apparatus according to claim 3 .

5. The control unit is further configured to cause the large-scale generative model to estimate a skill level of the user by providing a fifth prompt including the user input and an instruction to estimate the skill level.

5. An architecture proposal apparatus as claimed in claim 4.

6. The control unit is if the input for the query includes a response that the query is unanswerable, providing a sixth prompt including the query, the user's skill level with respect to designing the architecture, and a modification instruction, causing the large-scale generative model to generate a modified query; outputting the modified query to the user; and accepting input of a response to the revised inquiry from the user; 、 and further configured to perform updating the user requirements comprises updating the user requirements with the answers entered to the revised queries. The architecture proposal apparatus according to claim 1 .

7. The control unit is further configured to store the user's input, instructions for the large-scale generative model, and the generation result of the large-scale generative model in a chronological order as an interaction history. The architecture proposal apparatus according to claim 1 .

8. The control unit is outputting the generated proposed architecture to the user; and receiving input from the user specifying portions of the proposed architecture that will not be changed; and further configured to perform The re-proposal instruction includes an instruction indicating that the specified portion is not to be changed. The architecture proposal apparatus according to claim 1 .

9. The control unit is outputting the generated proposed architecture to the user; and receiving input from the user specifying portions of the proposed architecture to be modified; and further configured to perform The re-proposal instruction includes an instruction sentence indicating that the specified portion is to be changed. The architecture proposal apparatus according to claim 1 .

10. the control unit is further configured to cause the large-scale generative model to generate alternatives to the proposed architecture by providing a seventh prompt including the proposed architecture, the validation results, and instructions for generating alternatives; the third prompt further includes the generated alternatives. The architecture proposal apparatus according to claim 1 .

11. The control unit is providing an eighth prompt including the proposed architecture and a summary instruction to cause a large-scale generative model to generate a summary of the proposed architecture; and outputting the generated summary to the user; and further configured to perform The architecture proposal apparatus according to claim 1 .

12. the instruction to generate a query includes an instruction to generate a query item having a restricted input format for a response; The architecture proposal apparatus according to claim 1 .

13. The architecture proposal apparatus of claim 1 , wherein the query generation instructions comprise a directive that lists query items in order of priority.

14. the query generation instruction comprises an instruction for generating a query item in a form asking the user's request; The architecture proposal apparatus according to claim 1 .

15. The control unit is configured to manage the user requirements, the proposed architecture, the verification results, and the inquiry items as structured internal information. The architecture proposal apparatus according to claim 1 .

16. 1. A computer-implemented information processing method, comprising: accepting input of user requirements for an architecture relating to the information system from a user; generating a proposed architecture for the information system in a large-scale generative model upon providing a first prompt including the input user requirements and proposed instructions; causing the large-scale generative model to generate a validation result for the proposed architecture by providing a second prompt including the generated proposed architecture and a validation instruction; if there is a deficiency in the generated verification result, generating a query regarding the deficiency in the large-scale generative model by providing a third prompt including the proposed architecture, the verification result, and an instruction to generate a query; outputting the generated query to the user; accepting input of a response to the inquiry from the user; updating the user requirements with the input answers; causing the large-scale generative model to generate an updated proposed architecture by providing a fourth prompt including the updated user requirements, the proposed architecture, the validation results, the query, and a re-proposal instruction; refining the proposed architecture by repeating the steps of generating verification results for the proposed architecture, generating queries, outputting the queries, accepting input of responses to the queries, updating the user requirements, and generating the updated proposed architecture until no deficiencies exist; and outputting an improved version of the proposed architecture; Including, Information processing methods.

17. A program for causing a computer to execute an information processing method, The information processing method includes: accepting input of user requirements for an architecture relating to the information system from a user; generating a proposed architecture for the information system in a large-scale generative model upon providing a first prompt including the input user requirements and proposed instructions; causing the large-scale generative model to generate a validation result for the proposed architecture by providing a second prompt including the generated proposed architecture and a validation instruction; if there is a deficiency in the generated verification result, generating a query regarding the deficiency in the large-scale generative model by providing a third prompt including the proposed architecture, the verification result, and an instruction to generate a query; outputting the generated query to the user; accepting input of a response to the inquiry from the user; updating the user requirements with the input answers; causing the large-scale generative model to generate an updated proposed architecture by providing a fourth prompt including the updated user requirements, the proposed architecture, the validation results, the query, and a re-proposal instruction; generating a verification result of the proposed architecture, generating queries, outputting the queries, accepting input of responses to the queries, updating the user requirements, and refining the proposed architecture by iteratively generating the updated proposed architecture; and outputting an improved version of the proposed architecture; Including, program.

Images