Multi-model based drug management system, method, storage medium, and program product

Abstract

This application relates to the field of artificial intelligence technology, providing a multi-model-based drug management system, method, storage medium, and program product. The system includes: a drug management platform for processing drug management business logic and sending business requests outwards according to the business process; an intelligent agent for receiving business requests from the drug management platform, performing feature analysis on the requests, selecting and calling AI models, integrating the output results of the AI ​​models, and returning them to the drug management platform; and a model service platform for integrating, managing, and providing various types of AI models, receiving and responding to AI model call requests from the intelligent agent. In this application, the drug management platform and model service platform are decoupled and communicate through the intelligent agent, improving scalability and stability. The intelligent agent can analyze business requests, select suitable AI models, improve processing efficiency, and meet diverse business needs.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to multi-model-based drug management systems, methods, storage media, and program products. Background Technology

[0002] In the current field of drug regulation, existing drug management systems face many pressing problems that need to be addressed.

[0003] First, the drug administration process is complex, involving numerous review stages. Different business stages have different requirements for the appropriate AI models, varying significantly in model type, scale, and supplier. For example, drug quality testing may require high-precision image recognition models, while drug market demand analysis relies more heavily on big data prediction models. Furthermore, with rapid technological advancements and constantly evolving business needs, AI models are frequently updated and iterated, posing a significant challenge to existing systems in terms of model updates and adaptation.

[0004] Secondly, the AI ​​models used in the system mainly consist of a few vertical domain-specific models and most open-source models. Vertical domain-specific models offer high accuracy in specific business scenarios, but their development and maintenance costs are high. While open-source models offer advantages such as low cost and high customizability, they have certain shortcomings in terms of professionalism and stability. The coexistence of advantages and disadvantages among various models means that in practical applications, the model outputs still largely require manual review. This not only increases labor and time costs but is also susceptible to human factors, reducing the accuracy and efficiency of the review process. The existing drug management system is no longer able to meet the increasingly complex business needs. Summary of the Invention

[0005] This application provides a multi-model-based drug management system, method, storage medium, and program product that can effectively integrate various types of AI models to achieve efficient and collaborative drug management.

[0006] This application provides a multi-model-based drug management system, comprising: a drug management platform for processing drug management business logic and sending business requests outward according to the business process; an intelligent agent for receiving business requests from the drug management platform, performing feature analysis on the business requests, selecting and calling AI models, integrating the output results of the AI ​​models and returning them to the drug management platform; and a model service platform for integrating, managing and providing various types of AI models, receiving and responding to AI model calling requests from the intelligent agent; wherein the drug management platform and the model service platform are decoupled and do not directly interact with each other; the drug management platform and the model service platform establish communication through the intelligent agent.

[0007] According to the multi-model-based drug management system provided in this application, it also includes a support component, which is used to provide the intelligent agent with the technical support required for session management and model optimization. The intelligent agent is specifically used to: after receiving a business request from the drug management platform, identify the feature information contained in the business request; the feature information includes at least one of semantic features, data modality features, and application scenario features; based on the feature information, select one or more suitable target AI models from the AI ​​models managed by the model service platform; and collaboratively call the target AI models to process the business request.

[0008] According to the multi-model-based drug management system provided in this application, the intelligent agent is also used to: initiate a model information subscription to the model service platform to obtain a list of model information; and receive the latest model information pushed by the model service platform when the AI ​​model of the model service platform is updated, so as to ensure the real-time nature of the model information list.

[0009] According to the multi-model-based drug management system provided in this application, the supporting components include a session database; the intelligent agent is specifically used to: retrieve, create or update model session IDs in the session database based on the process ID of the business process; and use the model session IDs to call the model service platform to maintain context consistency of multiple model calls under the same business process.

[0010] According to the multi-model-based drug management system provided in this application, the intelligent agent is specifically used to: use the process ID as a query condition to search in the session database for whether there is a corresponding model session record; if there is a valid model session record, obtain the model session ID in the model session record to initiate subsequent model calls.

[0011] According to the multi-model-based drug management system provided in this application, the intelligent agent is specifically used to: use the process ID as a query condition to search in the session database for whether there is a corresponding model session record; if there is no valid model session record, generate a new model session ID, and create or update the association record between the process ID and the new model session ID in the session database, so as to initiate subsequent model calls using the new model session ID.

[0012] According to the multi-model-based drug management system provided in this application, the multi-model-based drug management system is also used to realize feedback-based model optimization, wherein: the drug management platform is also used to collect feedback data generated by manual review and send the feedback data to the intelligent agent; the intelligent agent is also used to process the feedback data into structured training data, save and record the storage address of the training data; the supporting components also include a feedback optimization component, which is used to trigger the model service platform to optimize the training of the corresponding AI model based on the training data.

[0013] This application also provides a multi-model-based drug management method. Using the aforementioned multi-model-based drug management system, the multi-model-based drug management method includes: receiving a business request from a drug management platform; selecting one or more suitable target AI models from a model information list based on the feature information of the business request; calling the target AI model to process the business request; and integrating the processing results of the target AI model and returning them to the drug management platform.

[0014] This application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multi-model-based drug management method as described above.

[0015] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the multi-model-based drug management method as described above.

[0016] This application provides a multi-model-based drug management system, method, storage medium, and program product. The system includes: a drug management platform for processing drug management business logic and sending business requests outwards according to the business process; an intelligent agent for receiving business requests from the drug management platform, performing feature analysis on the requests, selecting and calling AI models, integrating the output of the AI ​​models, and returning them to the drug management platform; and a model service platform for integrating, managing, and providing various types of AI models, receiving and responding to AI model call requests from the intelligent agent. The drug management platform and the model service platform are decoupled and do not directly interact with each other; communication between them is established through the intelligent agent. Through this method, the drug management platform and the model service platform are decoupled, and communication via the intelligent agent improves scalability and stability. The intelligent agent can analyze business requests and select suitable AI models, improving processing efficiency. Furthermore, supporting components are provided to offer session management and model optimization support, meeting diverse business needs. Attached Figure Description

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

[0018] Figure 1 This is a schematic diagram of the structure of a multi-model-based drug management system provided in an embodiment of this application.

[0019] Figure 2This is a schematic diagram of the overall technical architecture of the multi-model-based drug management system provided in the embodiments of this application.

[0020] Figure 3 This is a schematic diagram illustrating the interaction of an agent subscribing to model information provided in an embodiment of this application.

[0021] Figure 4 This is an interactive schematic diagram of the intelligent agent adaptation model provided in the embodiments of this application.

[0022] Figure 5 This is an interactive diagram illustrating process session consistency provided in the embodiments of this application.

[0023] Figure 6 This is an interactive diagram illustrating the collection and storage of manual review feedback data provided in an embodiment of this application.

[0024] Figure 7 This is an interactive schematic diagram of training and optimizing a model based on feedback data, provided in an embodiment of this application.

[0025] Figure 8 This is a flowchart illustrating the multi-model-based drug management method provided in the embodiments of this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the embodiments of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0028] This application provides a multi-model-based drug management system. Please refer to [link / reference]. Figure 1 , Figure 1This is a schematic diagram of the structure of a multi-model-based drug management system provided in an embodiment of this application. In this embodiment, the multi-model-based drug management system may include a drug management platform 110, an intelligent agent 120, a model service platform 130, and supporting components 140.

[0029] The drug management platform 110 is used to process drug management business logic and send business requests to external entities according to the business process.

[0030] Intelligent agent 120 is used to receive business requests from the drug management platform, perform feature analysis on the business requests, select and call AI models, integrate the output results of the AI ​​models, and return them to the drug management platform.

[0031] The model service platform 130 is used to integrate, manage and provide various types of AI models, and to receive and respond to AI model invocation requests from intelligent agents.

[0032] In this system, the drug management platform 110 is decoupled from the model service platform 130 and does not directly interact with data. The drug management platform 110 and the model service platform 130 communicate with each other through the intelligent agent 120.

[0033] In this embodiment, the multi-model-based drug management system adopts a modular design, dividing the system into a drug management platform 110, an intelligent agent 120, and a model service platform 130. Each component has a clear division of labor and collaborates through specific interaction methods to achieve intelligent processing of drug management operations.

[0034] Optionally, the multi-model-based drug management system may also include a support component 140, which provides the agent with the technical support required for session management and model optimization.

[0035] Specifically, the drug management platform 110, as the core of the system's business, can make judgments on different business scenarios based on preset business processes and send out corresponding business requests.

[0036] After receiving a business request from the drug management platform, the intelligent agent 120 can perform feature analysis on the request to understand the specific content and requirements of the business request and extract key feature information.

[0037] Based on the feature analysis results, Agent 120 can select the most suitable AI model from multiple available options. Different business requests may require different types of AI models. Agent 120 calls the selected AI model to perform calculations and integrates the outputs of these models. Finally, the integrated result is returned to the drug management platform for subsequent business decisions.

[0038] The model service platform 130 is used to integrate and manage various types of AI models, including model storage, updates, and version control. The model service platform 130 can provide a unified interface to receive AI model invocation requests from intelligent agents, instantiate and run the corresponding AI models according to the requests, and return the model's calculation results to the intelligent agents.

[0039] Support component 140 provides essential technical support for agent 120. In terms of session management, support component 140 ensures the orderly interaction between the agent and the drug management platform 110 and model service platform 130, records session information, and guarantees data consistency and integrity. In terms of model optimization, support component 140 collects model runtime data, analyzes model performance indicators, and provides a basis for model adjustment and optimization.

[0040] The supporting components enable the system to easily manage sessions and optimize models, providing technical support for system expansion. As business grows and changes, session management and model optimization strategies can be adjusted at any time.

[0041] It should also be noted that in this embodiment, the drug management platform 110 and the model service platform 130 are decoupled, meaning that there is no direct data interaction between them. This design avoids a strong dependency between the two platforms. Communication between them is established through intelligent agents. The drug management platform 110 sends business requests to the intelligent agent 120, which processes the requests and then requests the corresponding AI model services from the model service platform 130, before feeding the results back to the drug management platform.

[0042] As a result of the decoupling of the drug management platform and the model service platform, adding new business functions or introducing new AI models will not cause large-scale changes to the existing system architecture. The decoupling design reduces the coupling between different parts of the system, and the failure or abnormality of one component will not easily affect other components. The intelligent agent plays a buffering and coordinating role in the system. It can preprocess and schedule business requests, avoid direct conflicts between the drug management platform and the model service platform, and improve the overall stability of the system.

[0043] In this embodiment, the intelligent agent can perform feature analysis on business requests and select the most suitable AI model for processing. This targeted model selection can fully leverage the performance of AI models, reduce unnecessary computation and resource waste, and improve the speed and accuracy of business processing. The model service platform integrates various types of AI models, allowing the intelligent agent to flexibly select the appropriate model based on different business requests. This enables the system to adapt to various complex scenarios in drug management, meet diverse business needs, and provide a comprehensive intelligent solution for drug management.

[0044] Drug management platforms come in various types, including drug regulatory business platforms that specifically serve drug regulatory business. The following explanation will take a drug management platform applied to drug regulatory business as an example.

[0045] Drug regulatory services can cover six main stages: drug research and development, drug registration, drug production, drug distribution, drug use, and post-marketing supervision.

[0046] Please see Figure 2 , Figure 2 This is a schematic diagram of the overall technical architecture of a multi-model-based drug management system provided in this embodiment. In this embodiment, the overall architecture adopts a layered decoupling design, using intelligent agents to decouple the drug management platform from the AI ​​model, and relying on supporting components to complete the intelligent scheduling, capability iteration, and optimization of the model. The descriptions of each part are as follows: 1. Drug Management Platform: Focusing on the regulatory needs of the entire drug lifecycle, this platform provides full-scenario business applications covering all stages of regulatory processes. It undertakes core responsibilities such as business logic operations, user interface presentation, and full lifecycle management of business data. Instead of directly calling specific AI models, it implements various intelligent functions through interaction with intelligent agents. 2. Intelligent Agents: Based on the core needs of AI application scenarios in drug regulation, intelligent agents can be divided into four categories: ① Approval Assistance Intelligent Agent: Focuses on the review, evaluation and approval management of registration application materials for drugs, medical devices and cosmetics, supporting scenarios such as formal review, assisted evaluation and approval organization.

[0047] ② Regulatory Enforcement Intelligent Agent: Responsible for supporting the entire process of daily regulatory activities, covering remote monitoring, on-site inspection, sampling and pharmacovigilance, etc., and supporting AI application scenarios such as remote supervision, on-site supervision, and pharmacovigilance.

[0048] ③ Public Service Intelligent Agent: Provides government support to the public and enterprises, such as consultation and response, application assistance, and age-friendly services, to improve service experience and accessibility, and support AI application scenarios such as business processing guidance, consultation and answering, and age-friendly modification of instruction manuals.

[0049] ④ Decision Support Intelligent Agent: Provides support for regulatory decision-making, including data query, in-depth analysis, trend prediction, and solution research, optimizes the construction of risk management system, and supports AI application scenarios such as business data query, data analysis and prediction, and risk management assessment.

[0050] 3. Supporting Components: These provide the underlying technical support for the implementation of innovative system functions, specifically including two types of core components: ①Session Database: Used to store the session context of the same drug regulatory business process, specifically covering key data such as business process documents associated with the session (such as approval number, batch number), model representation, model session identifier, and manual review feedback records (such as correction opinions, regulatory basis).

[0051] ② Feedback optimization component: Based on manually reviewed feedback data, the model's capabilities are iteratively optimized through a closed-loop mechanism of "sample selection - incremental training - effect verification".

[0052] 4. Model Service Platform: This platform integrates various AI models, including large language models, computer vision models, time series prediction models, and professional analysis models for the drug regulatory field. It supports the entire lifecycle management of models, including import and storage, version control, training and fine-tuning, evaluation and validation, deployment and launch, monitoring and maintenance. Stable model services are provided externally through standardized API interfaces.

[0053] In some embodiments, the intelligent agent is specifically used to: after receiving a business request from the drug management platform, identify the feature information contained in the business request; the feature information includes at least one of semantic features, data modality features, and application scenario features; based on the feature information, select one or more suitable target AI models from the AI ​​models managed by the model service platform; and collaboratively invoke the target AI models to process the business request.

[0054] In this embodiment, after receiving a business request from the drug management platform, the intelligent agent can perform natural language processing on the text information in the request. For example, it can analyze the vocabulary, grammar, and semantics of the request statement to understand the specific meaning of the business request.

[0055] Business requests can contain different types of feature information, and the intelligent agent needs to identify this data in order to select the appropriate AI model. For example, the intelligent agent needs to determine the specific application scenario of the business request. Drug management business involves multiple scenarios, each with different business requirements and data characteristics. By identifying application scenario features, the intelligent agent can better match the appropriate AI model.

[0056] Based on the identified feature information, the intelligent agent filters among numerous AI models managed by the model service platform. It selects target AI models based on the matching degree between different feature information and model capabilities. Once one or more target AI models are selected, the intelligent agent is responsible for collaboratively calling these models to process business requests.

[0057] Specifically, if only one target AI model is selected, the agent directly calls that model to process the business request. If multiple target AI models are selected, the agent needs to coordinate the working order and data interaction of these models. The agent processes and distributes the data of the business request according to the input requirements of the models, collects the output results of each model, and integrates and comprehensively analyzes them. In this embodiment, by identifying and analyzing various feature information in the business request, the agent can more accurately select the appropriate target AI model; considering various feature information enables the agent to adapt to various complex business scenarios in drug management. Business requests in different scenarios have different characteristics. By identifying application scenario characteristics, the agent can select the most suitable model for each scenario; in addition, collaboratively calling multiple target AI models can fully leverage the advantages of different models and improve the overall processing capability of the system. Different AI models have their own strengths in processing different types of data and problems. Through collaborative calling by the agent, these strengths can be combined. Collaborative work can achieve more comprehensive and accurate drug management.

[0058] In some embodiments, the intelligent agent is also configured to: initiate a model information subscription to the model service platform to obtain a list of model information; and receive the latest model information pushed by the model service platform when the AI ​​model on the model service platform is updated, so as to ensure the real-time nature of the model information list.

[0059] In this embodiment, the agent can proactively initiate a model information subscription request to the model service platform. In the subscription request, the agent can specify the specific content to subscribe to, such as the model's name, type, version number, update time, etc., or it can choose to subscribe to information for all models.

[0060] After receiving a subscription request from an agent, the model service platform filters and organizes relevant model information based on the request content, and provides this information to the agent in the form of a list. The model information list contains detailed information about each AI model, which the agent can use to subsequently select the appropriate AI model to handle business requests.

[0061] In addition, the model service platform can continuously monitor the status of the AI ​​models it manages. When an AI model is updated, such as when parameters are adjusted, the structure is optimized, or new training data is added, the model service platform can capture these update events.

[0062] When a model update is detected, the model service platform can push the latest model information to agents that have subscribed to the model information. After receiving the pushed latest model information, the agent can update its local model information list to ensure that the list contains only the latest model information.

[0063] Alternatively, the push mechanism can also be updated at fixed times or on fixed cycles, such as monthly updates or weekly updates.

[0064] Optionally, the push mechanism can be real-time to ensure that the agent can obtain the latest state of the model in a timely manner.

[0065] The updated list of model information enables the agent to make decisions based on the latest model status when selecting a target AI model. When new AI models are added to the model service platform or existing models are updated, the agent is promptly informed of these changes. This allows the system to quickly adapt to new business needs and data characteristics, providing more effective support for drug management and enhancing the system's flexibility and adaptability.

[0066] Furthermore, the model information subscription and push mechanism enables automated model information updates, reducing the need for manual intervention. In traditional systems, it may be necessary to manually check the model's update status periodically and update the relevant model information manually. However, in the system of this embodiment, the agent can automatically obtain the latest model information, greatly reducing the cost and workload of manual maintenance.

[0067] To decouple the drug management platform from the model, this application proposes to use intelligent agents to achieve intelligent adaptation of the model in the business process.

[0068] First, each agent should be aware of the model information integrated into the current model service platform. This proposal suggests a technical solution where agents subscribe to model information from the model service platform. Please refer to [link / reference]. Figure 3 , Figure 3 This is a schematic diagram illustrating the interaction of an agent subscribing to model information provided in an embodiment of this application.

[0069] 1. When an agent initiates a model subscription request to the model service platform, it must include the filtering conditions for the required models, including but not limited to model type (such as large language models, multimodal models, etc.), keywords, parameter size, and model provider. If the filtering conditions are empty, all model information on the platform will be subscribed to by default. Agents with different functions can subscribe to corresponding model information based on the business characteristics of their services. The specific correspondence can be seen in the table below:

[0070] 2. After receiving a subscription request, the model service platform will filter models that meet the criteria based on the filtering conditions and return a list of model information, including key attributes such as model type, name, provider, and number of parameters. 3. Due to the dynamic changes (including addition, deletion and upgrade) of the models integrated by the model service platform, when a model is updated and meets the subscription conditions of a certain agent, the platform will automatically push the updated information to all agents that have subscribed to the model to ensure the real-time and accuracy of the model information held by the agents.

[0071] Please see Figure 4 , Figure 4 This is an interactive schematic diagram of the intelligent agent adaptation model provided in the embodiments of this application.

[0072] 1. When creating a business process in the workflow orchestration module of the drug management platform, you can configure whether to enable intelligent functions for each node. If enabled, you need to specify the corresponding intelligent agent. For example, in the drug registration process, both the formal review and technical review nodes can call the approval assistance intelligent agent.

[0073] 2. During process execution, after receiving files and information submitted by the user, the intelligent agent can automatically identify multi-dimensional features and, based on its built-in model adaptation algorithm, intelligently select an applicable model from the list of model information subscribed to by the model service platform. The identified features may include: ① Semantic features: Extract keywords from user input information, file names and content to identify user intent.

[0074] ② Data modality characteristics: Identify the modality type of user-submitted data, such as text, images, scanned electronic documents, videos, structured data, etc.

[0075] ③ Application scenario characteristics: including information such as the application system that initiated the business process, the process name, and the current node, used to determine the current business scenario.

[0076] For example, in the scenario of determining the level of change in drug registration, after receiving the multi-format application materials submitted by the drug management platform, the intelligent agent automatically identifies and extracts keywords such as "change in production process", data modal features (such as parsing text, images and scanned documents) and application scenario features (locating the change management process under the review system), and intelligently selects and collaboratively calls the large language model (analyzing text and regulations), the computer vision model (processing process images) and the predictive analysis model (assessing risks) based on the built-in algorithm.

[0077] 3. The agent initiates a model session request to the model service platform. A single session can include calls to multiple models. After integrating the results returned by the models, the agent returns the final result to the drug management platform.

[0078] In the field of drug regulation, it is impossible to solve all problems with a single model. Therefore, this embodiment can deploy a variety of different types of AI models to adapt to diverse business processes and application scenarios.

[0079] It should be noted that drug regulatory processes are complex and involve numerous review stages. The types, scales, and suppliers of models required for each business node vary significantly, and AI models are frequently updated and iterated. Relying on drug regulatory system administrators to manually configure and maintain the corresponding models for each stage is impractical. Therefore, this application's embodiments establish an intelligent AI model adaptation mechanism, enabling customers and drug regulatory systems to obtain the optimal intelligent service experience without needing to concern themselves with the specific models used in the backend.

[0080] In related technologies, the AI ​​models used by the system typically consist of a few vertical domain-specific models and a majority of open-source models, each with its own advantages and disadvantages. In practical applications, the model outputs mostly still require manual review to ensure their scientific validity and accuracy. However, the experience accumulated through manual review has not been effectively fed back to the backend model, resulting in the model output failing to be continuously optimized, falling into a cycle of "repeated manual corrections," and wasting human resources.

[0081] Based on this, this application proposes a self-optimizing multi-model intelligent drug supervision system framework. This framework decouples the drug management platform from the backend model through multiple functional intelligent agents, and on this basis, achieves context-aware reasoning optimization through a process and model session consistency mechanism. Furthermore, it utilizes human feedback data to train and optimize the backend model.

[0082] In some embodiments, the supporting components include a session database; the agent is specifically used to: retrieve, create or update a model session ID in the session database based on the process ID of the business process; and use the model session ID to call the model service platform to maintain context consistency for multiple model calls under the same business process.

[0083] The session database is a crucial component of the supporting system, primarily used to store and manage information related to model sessions. In the drug management system, each business process has its unique process ID, and each session associated with a model call has a corresponding model session ID. The session database acts as an information repository, providing agents with the functionality to store and retrieve this critical information.

[0084] After receiving a business request from the drug management platform, the agent retrieves it from the session database based on the process ID of the business process. If there is already a related model call session for that business process, the agent can find the corresponding model session ID.

[0085] If no model session ID corresponding to the current business process is found in the session database, the agent will create a new model session ID. The newly created model session ID will be stored in the session database and associated with the corresponding process ID.

[0086] During the execution of a business process, the model session ID may need to be updated. This could be because the state of the model session has changed, or because new session information needs to be recorded. For example, when a model call session completes part of a task and needs to move to the next stage, the agent may update the session state information associated with the model session ID to reflect the current business progress.

[0087] The agent uses the model session ID to invoke the model service platform, ensuring contextual consistency across multiple model calls within the same business process. In pharmaceutical management, a single business process may require multiple calls to different AI models. The model service platform can identify that these calls belong to the same business process, thus maintaining contextual consistency when processing these requests. It can leverage the results and relevant information from previous model calls to provide more accurate input for subsequent model calls, avoiding information loss and redundant calculations.

[0088] In some embodiments, the agent is specifically used to: use the process ID as a query condition to search the session database for whether a corresponding model session record exists; if a valid model session record exists, obtain the model session ID in the model session record to initiate subsequent model calls.

[0089] The agent uses a session database to manage and track the association between business processes and model sessions. The session database stores model session records for each business process, including the model session ID and other relevant session information. The agent queries the database using the process ID, leveraging the database's indexing mechanism to quickly locate the corresponding record and determine if a valid model session record exists.

[0090] After obtaining the process ID of the business process, the intelligent agent sends it to the session database as a query condition. Upon receiving the query request, the session database searches for the process ID in its stored records.

[0091] Specifically, the database can use a pre-built index structure to quickly locate the record associated with the process ID. If the corresponding record exists in the database, its validity will be further checked. Validity checks may include checking the record's status fields, such as whether it has expired or been completed, to ensure that the record is usable.

[0092] If a valid model session record exists in the session database, the agent will extract the model session ID from that record. This model session ID is a key identifier for subsequent interactions with the model service platform, representing a specific model call session within that business process. The agent uses the obtained model session ID to initiate subsequent model call requests to the model service platform. Based on this model session ID, the model service platform can identify the business process context to which the request belongs, thereby maintaining the continuity and consistency between multiple model calls.

[0093] In some embodiments, the agent is specifically used to: use the process ID as a query condition to search the session database for whether a corresponding model session record exists; if no valid model session record exists, generate a new model session ID, and create or update an association record between the process ID and the new model session ID in the session database, so as to initiate subsequent model calls using the new model session ID.

[0094] In this embodiment, if no valid model session record is found in the session database, the agent will perform the following operations: generate a new model session ID and create or update the associated record.

[0095] Specifically, the agent uses a specific algorithm to generate a unique model session ID. This algorithm typically considers factors such as timestamps and random numbers to ensure the uniqueness of the generated ID. For example, a new model session ID can be created by combining the current timestamp with a randomly generated number. The agent associates the newly generated model session ID with the current process ID and creates a new record in the session database. If the process ID already exists in the database but the corresponding record is invalid, the agent may update the record and associate it with the new model session ID. In this way, the session database records the correspondence between the business process and the new model session ID.

[0096] The agent uses the newly generated model session ID to initiate subsequent model call requests to the model service platform. Upon receiving the request, the model service platform will process the model call for this business process based on this model session ID, ensuring that context consistency is maintained in subsequent processing.

[0097] Please see Figure 5 , Figure 5 This is an interactive diagram illustrating process session consistency provided in the embodiments of this application.

[0098] For example, to ensure contextual consistency in intelligent approval processes for pharmaceuticals and cosmetics within the same approval workflow, this embodiment proposes a dedicated session management mechanism to ensure that the model can perform accurate reasoning based on complete contextual information. This mechanism first establishes a session database table, the structure of which is shown below:

[0099] The specific process is as follows: 1. After receiving a request from the drug management platform, the intelligent agent first selects the model that is suitable for the current process node.

[0100] 2. The agent uses the process ID as the query condition to search the session database for a session record corresponding to the process. If it exists, proceed to step 3; otherwise, proceed to step 5.

[0101] 3. The agent retrieves the session ID from the retrieved session records and uses it to initiate a business request to the model service platform. If the return message indicates that the session has expired, proceed to step 4; otherwise, continue executing subsequent business based on the existing session, and the process ends.

[0102] 4. The agent accesses the model service platform with an empty session ID, which generates a new session ID. The agent uses the new ID to update the session ID field in the original session record and continues to use this new ID in subsequent interactions, at which point the process ends.

[0103] 5. The agent accesses the model service platform with an empty session ID, obtains a new session ID, inserts a new record in the session database, saves the current process ID and the new session ID, and the process ends.

[0104] In the model service platform, session management is the core mechanism for achieving efficient and controllable interaction. Its systematic design significantly improves the quality of model inference and user experience. This mechanism mainly embodies the following four key characteristics: Independence: Each session has its own independent context and state storage, achieving complete data isolation between different sessions, effectively preventing information leakage and mutual interference, thereby ensuring privacy and security in multi-user concurrent scenarios.

[0105] Coherence: Within the same session, the system can connect multiple user requests in context based on historical interaction content. By automatically associating dialogue sequences, the model can more accurately understand user intent and generate logically clear and coherent responses.

[0106] Consistency: The system maintains the consistency of the context state throughout the entire session lifecycle, ensuring the reliability and predictability of the model output regardless of changes in request order or frequency.

[0107] Lifecycle management: Provides mechanisms such as proactive session creation, automatic recycling upon idle timeout, and manual destruction. With reasonable resource allocation and release strategies, it improves service availability while avoiding resource waste and ensuring long-term stable system operation.

[0108] Based on the above characteristics, session management brings many benefits to model inference: it enhances contextual understanding, thereby improving the accuracy and relevance of multi-turn dialogues; it supports personalized interaction, providing users with more accurate services based on session history; it optimizes system resource utilization, reducing the overhead of repetitive calculations and redundant requests; and it enhances the maintainability and controllability of services, facilitating system fault diagnosis and operation auditing.

[0109] In some embodiments, the multi-model-based drug management system is further used to implement feedback-based model optimization, wherein: the drug management platform is further used to collect feedback data generated by manual review and send the feedback data to the intelligent agent; the intelligent agent is further used to process the feedback data into structured training data, save and record the storage address of the training data; the supporting components also include a feedback optimization component, which is used to trigger the model service platform to optimize the training of the corresponding AI model based on the training data.

[0110] This embodiment describes the process of implementing feedback-based model optimization in a multi-model-based drug management system.

[0111] In drug administration, manual review is a crucial step. The drug administration platform collects feedback data generated during this process. This data may include reviewers' suggestions for correcting the model's output and their assessments of data accuracy. After collecting the feedback data, the drug administration platform sends it to the intelligent agent, providing raw materials for subsequent model optimization.

[0112] After receiving feedback data, the agent processes this unstructured or semi-structured feedback data into structured training data. The processed training data is saved to a designated storage location, and the agent records the storage address of this training data. This facilitates the subsequent feedback optimization component in quickly locating and using this data.

[0113] The feedback optimization component retrieves training data from the training data storage address provided by the agent, and then triggers the model service platform to optimize the training of the corresponding AI model. This allows the model service platform to retrain the inventory management model, adjusting its parameters and structure to improve prediction accuracy.

[0114] In summary, this embodiment, by continuously collecting feedback data from manual review and using it for model optimization, enables the AI ​​model to better adapt to the needs of actual drug management operations, improving the accuracy of model prediction and decision-making. The feedback data from manual review provides valuable information for model optimization, while the optimization of the model provides more accurate references for manual review, achieving positive interaction and collaboration between humans and machines, and improving the overall efficiency and quality of drug management.

[0115] Please see Figure 6-7, Figure 6 This is an interactive diagram illustrating the collection and saving of manual review feedback data provided in an embodiment of this application. Figure 7 This is an interactive schematic diagram of training and optimizing a model based on feedback data, provided in an embodiment of this application.

[0116] The intelligent agent establishes a feedback learning loop mechanism, which can transform the professional judgments of human review into training data that the model can learn from, thereby continuously improving the accuracy and practicality of the model.

[0117] Combination Figure 6 The process for collecting and saving feedback data through manual review is as follows: 1. When a human reviewer modifies or overturns the model's output, the drug management platform sends human feedback data to the agent corresponding to that process node. This feedback includes the process ID, review basis, correction suggestions, and the final review result. For example, in the scenario of determining the change level for drug registration, if a human reviewer modifies the change level determined by the model, the platform will record and provide feedback on the results before and after the modification, the applicable laws and regulations, and the professional review reasons.

[0118] 2. The agent analyzes the feedback data, identifies the model that needs correction based on data characteristics, and transforms the feedback information into structured and standardized training data using natural language processing and technical terminology extraction techniques. This data is stored in a specified path in the form of datasets or files, with directories generated according to predetermined naming rules, such as "process ID + model ID + current date and time".

[0119] 3. The agent uses the process ID as the search condition to find the corresponding session record in the session database and adds the storage address of the feedback data (i.e. the directory path generated in step 2) to the list of storage addresses for human feedback information of that record.

[0120] Combination Figure 7 Based on the feedback data saved in the above process, the feedback optimization component will perform data sampling and model training optimization. The specific process is as follows: 1. The feedback optimization component periodically extracts incremental data from the session database based on the last update time in the session records. For example, at midnight each day, it extracts the session records updated the previous day, reads the list of feedback data storage addresses, accesses the corresponding addresses, and retrieves the feedback data.

[0121] 2. The feedback optimization component samples incremental feedback data and supplements it to the training dataset of the corresponding model. Different feedback processing strategies are adopted for different types of models. For example, for deep learning-based neural network models, feedback data is mainly used to fine-tune model parameters; for rule-based knowledge graph models, feedback data is used to optimize rule weights and supplement rule logic.

[0122] 3. The feedback optimization component dynamically adjusts the optimization frequency of each model based on factors such as model type, feedback data volume, and business impact. It then calls the model service platform's interface during periods of low system load (e.g., nighttime) to trigger model optimization training. The model service platform intelligently schedules training tasks based on model priority, resource requirements, and estimated training time to ensure no impact on daily business operations. After training, the new model version undergoes verification and testing. Once performance improvements are confirmed, it is gradually deployed to the production environment for smooth updates. The feedback optimization methods for different model types are shown in the table below:

[0123] This application also provides a multi-model-based drug management method, using the aforementioned multi-model-based drug management system.

[0124] Please see Figure 8 , Figure 8 This is a flowchart illustrating a multi-model-based drug management method provided in an embodiment of this application. In this embodiment, the multi-model-based drug management method may include steps S810 to S840, each step as follows: S810: Receives business requests from the drug management platform.

[0125] S820: Based on the feature information of the business request, select one or more suitable target AI models from the model information list.

[0126] S830: Call the target AI model to process business requests.

[0127] S840: After integrating the processing results of the target AI model, return them to the drug management platform.

[0128] The multi-model-based drug management method provided in this application significantly improves business processing efficiency. By automatically selecting and calling suitable AI models, it reduces manual intervention, enabling rapid processing of various business requests in drug management. Simultaneously, this method enhances decision-making accuracy by integrating the processing results of multiple AI models, combining the advantages of different models to provide a more comprehensive and accurate basis for drug management decisions. Furthermore, the decoupling of the drug management platform and model service platform within the system provides strong flexibility and scalability, allowing for dynamic selection of appropriate AI models based on different business requests, thus adapting to diverse drug management business needs.

[0129] In summary, the multi-model-based drug management system and corresponding method provided in this application can decouple the drug management platform and the model service platform, preventing direct interaction between the two. It can automatically select and call the most suitable AI model based on the characteristics of drug regulatory business requests, achieving intelligent scheduling and resource optimization. By maintaining the consistency of the session context within the same business process, the model can perform inference based on complete session information, significantly improving the rationality and accuracy of the output results. Furthermore, it can automatically train and optimize the backend model using feedback data generated during manual review, thereby achieving continuous self-improvement and performance enhancement of the entire system.

[0130] The technical solution proposed in this application significantly improves the accuracy, efficiency, and adaptability of drug regulation through multi-model collaboration and intelligent adaptation mechanisms. On the one hand, the system relies on a multi-agent architecture to decouple the business platform from the model service, enabling dynamic selection of the optimal model based on business characteristics and ensuring high-precision processing in professional scenarios. On the other hand, the system introduces process and session consistency technology, utilizes contextual reasoning to optimize the rationality of output, and continuously optimizes model performance through a closed-loop data feedback mechanism. This breaks the dilemma of repetitive manual review, realizes the transformation of the regulatory model from passive response to proactive prediction, and comprehensively improves the level of intelligent regulation.

[0131] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the multi-model-based drug management method provided by the above methods. The steps and principles of the method have been described in detail in the above methods and will not be repeated here.

[0132] In another aspect, this application also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform the multi-model-based drug management methods provided by the above methods. The steps and principles of these methods have been described in detail in the above methods and will not be repeated here.

[0133] Non-transitory computer-readable storage media can be any available medium or data storage device that can be accessed by a processor, including but not limited to magnetic storage (e.g., floppy disks, hard disks, magnetic tapes, magneto-optical disks (MOs), etc.), optical storage (e.g., CDs, DVDs, BDs, HVDs, etc.), and semiconductor storage (e.g., ROMs, EPROMs, EEPROMs, non-volatile memory (NAND flash), solid-state drives (SSDs)).

[0134] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0135] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A multi-model-based drug management system, characterized in that, include: The drug management platform is used to process drug management business logic and send business requests to external entities according to the business process. An intelligent agent is used to receive business requests from the drug management platform, perform feature analysis on the business requests, select and call AI models, integrate the output results of the AI ​​models, and return them to the drug management platform. The model service platform is used to integrate, manage, and provide various types of AI models, and to receive and respond to AI model invocation requests from the intelligent agents. The drug management platform is decoupled from the model service platform and does not directly interact with data. The drug management platform and the model service platform establish communication through the intelligent agent.

2. The multi-model-based drug management system according to claim 1, characterized in that, It also includes support components, which provide the agent with the technical support required for session management and model optimization; The intelligent agent is specifically used for: Upon receiving a business request from the drug management platform, the system identifies the feature information contained in the business request; the feature information includes at least one of semantic features, data modality features, and application scenario features. Based on the aforementioned feature information, one or more suitable target AI models are selected from the AI ​​models managed by the model service platform. The target AI model is collaboratively invoked to process the business request.

3. The multi-model-based drug management system according to claim 2, characterized in that, The intelligent agent is also used for: Initiate a model information subscription to the model service platform to obtain a list of model information; When the AI ​​model on the model service platform is updated, the latest model information pushed by the model service platform is received to ensure the real-time nature of the model information list.

4. The multi-model-based drug management system according to claim 2, characterized in that, The supporting components include a session database; the intelligent agent is specifically used for: Based on the process ID of the business process, retrieve, create, or update the model session ID in the session database; The model service platform is invoked using the model session ID to maintain context consistency across multiple model invocations within the same business process.

5. The multi-model-based drug management system according to claim 4, characterized in that, The intelligent agent is specifically used for: Using the process ID as the query condition, search the session database to see if a corresponding model session record exists; If a valid model session record exists, the model session ID in the model session record is obtained to initiate subsequent model calls.

6. The multi-model-based drug management system according to claim 4, characterized in that, The intelligent agent is specifically used for: Using the process ID as the query condition, search the session database to see if a corresponding model session record exists; If no valid model session record exists, a new model session ID is generated, and an association record between the process ID and the new model session ID is created or updated in the session database to initiate subsequent model calls using the new model session ID.

7. The multi-model-based drug management system according to any one of claims 2 to 6, characterized in that, The multi-model-based drug management system is also used to implement feedback-based model optimization, wherein: The drug management platform is also used to collect feedback data generated by manual review and send the feedback data to the intelligent agent; The intelligent agent is also used to process the feedback data into structured training data, and save and record the storage address of the training data; The support component also includes a feedback optimization component, which is used to trigger the model service platform to optimize the training of the corresponding AI model based on the training data.

8. A multi-model-based drug management method, characterized in that, Using the multi-model-based drug management system as described in any one of claims 1 to 7, the multi-model-based drug management method includes: Receive business requests from the drug management platform; Based on the feature information of the business request, select one or more suitable target AI models from the model information list; The target AI model is invoked to process the business request; The processing results of the target AI model are integrated and returned to the drug management platform.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the multi-model-based drug management method as described in claim 8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the multi-model-based drug management method as described in claim 8.

Images