Systems and methods for context-aware multi-module routing and dynamic workflow execution

Abstract

This disclosure relates to methods and systems for assisting users in performing context-aware multi-module routing and dynamic workflow execution in an industrial environment. The system includes a master agent and multiple machine learning (ML) modules. The master agent is configured to receive input and analyze it to identify one or more tasks to satisfy one or more user requirements. The master agent identifies one or more ML modules suitable for performing one or more tasks from among the multiple ML modules based on at least one of the following: historical usage patterns, the context of one or more tasks, and one or more keywords associated with one or more tasks. The master agent also generates and executes dynamic workflows for sequences of invoking the identified one or more ML modules. The master agent further generates a final result based on an aggregation of intermediate results generated by the one or more ML modules. The final result is provided as a response to the input to assist the user.

Description

Technical Field

[0001] This disclosure generally relates to artificial intelligence (AI) driven co-pilots. More specifically, this disclosure relates to methods and multi-agent co-pilot systems for assisting users by meeting user requirements relevant to industrial environments. Background Technology

[0002] With the rapid adoption of AI-driven co-pilots in various industrial applications, there is a need to effectively orchestrate multiple machine learning models to handle complex user requests. Traditional systems rely on static workflows and rigid module selection processes, which lead to inefficiencies when dealing with dynamic and evolving tasks. Current AI / ML integration frameworks face significant challenges in managing the dynamic invocation of multiple models and modules based on context-based user prompts. These systems often lack the flexibility to effectively integrate various AI / ML models, Large Language Models (LLM), and Industrial Internet of Things (IoT) specific models. Additionally, existing solutions do not provide a centralized decision-making mechanism capable of dynamically ordering and managing data flows across different modules, resulting in inefficiencies and limited scalability.

[0003] Therefore, there is a need for technologies that can overcome these limitations and facilitate seamless integration and communication of multiple LLM, AI / ML models, and other first-principles models in the IoT field.

[0004] The information disclosed in the background section of this disclosure is only intended to enhance the understanding of the overall background of the invention and should not be construed as an admission or any implication that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] In one embodiment, this disclosure discloses a system for assisting users in performing context-aware multi-module routing and dynamic workflow execution in an industrial environment. The system includes a master agent and multiple machine learning (ML) modules communicatively coupled to the master agent. The master agent is configured to receive input including one or more user requests related to the industrial environment. The master agent analyzes the input to identify one or more tasks to satisfy the one or more user requests, and identifies one or more ML modules suitable for performing the one or more tasks from among the multiple ML modules based on at least one of the following: historical usage patterns, the context of the one or more tasks, and one or more keywords associated with the one or more tasks. The master agent generates a dynamic workflow for invoking the identified one or more ML modules in sequence, and executes the dynamic workflow by invoking the identified one or more ML modules according to the sequence. Each of the identified one or more ML modules is selected to perform at least one of the one or more tasks and generate intermediate results. The master agent also generates a final result based on the aggregation of intermediate results generated by the one or more ML modules through the execution of the dynamic workflow, wherein the final result is provided as a response to input to assist the user. The dynamic workflow is adjustable based on the real-time results of the execution of the dynamic workflow.

[0006] In one embodiment, this disclosure discloses a method for assisting users in an industrial environment with context-aware multi-module routing and dynamic workflow execution. The method includes receiving input from a master agent, comprising one or more user requests related to the industrial environment. The method further illustrates that the master agent analyzes the input to identify one or more tasks to satisfy the one or more user requests, and identifies one or more ML modules suitable for performing the one or more tasks from a plurality of machine learning (ML) modules based on at least one of the following: historical usage patterns, the context of the one or more tasks, and one or more keywords associated with the one or more tasks. The method further illustrates that the master agent generates a dynamic workflow for invoking a sequence of the identified one or more ML modules. Furthermore, the method illustrates that the master agent executes the dynamic workflow by invoking the identified one or more ML modules according to the sequence. Each of the identified one or more ML modules is selected to perform at least one of the one or more tasks and generate intermediate results. The method further illustrates that the master agent generates a final result based on the aggregation of the intermediate results from the identified one or more ML modules, wherein the final result is provided as a response to input to assist the user. The dynamic workflow is adjustable based on the real-time results of the execution of the dynamic workflow.

[0007] The foregoing overview is illustrative only and is not intended to be limiting in any way. Other aspects, embodiments, and features will become apparent from the accompanying drawings and the following detailed description, in addition to the illustrative aspects, embodiments, and features described above. Attached Figure Description

[0008] The novel features and characteristics of this disclosure are set forth in the appended claims. However, the disclosure itself, as well as its preferred modes of use, further objects and advantages, will be best understood by reading in conjunction with the accompanying drawings and with reference to the following detailed description of illustrative embodiments. One or more embodiments will now be described by way of example only with reference to the accompanying drawings, wherein like reference numerals denote like elements, and in the drawings:

[0009] Figure 1 The illustration shows an exemplary environment according to some embodiments of the present disclosure for assisting a user in meeting user requirements related to an industrial environment;

[0010] Figure 2 The figure illustrates a detailed block diagram of a system according to some embodiments of the present disclosure for assisting users in meeting user requirements related to industrial environments;

[0011] Figure 3 The illustrations show exemplary diagrams of systems according to some embodiments of the present disclosure for assisting users in meeting user requirements related to industrial environments; and

[0012] Figure 4 The illustrations depict methods for assisting users in meeting user requirements related to industrial environments, according to some embodiments of the present disclosure.

[0013] Those skilled in the art will understand that any block diagram herein embodies a conceptual view of an illustrative system of principles of the subject matter. Similarly, it will be understood that any flowchart, schematic diagram, state transition diagram, pseudocode, etc., represents various processes that can be substantially represented in a computer-readable medium and executed by a computer or processor, whether or not such computer or processor is explicitly shown. Detailed Implementation

[0014] In this document, the word "exemplary" is used herein to mean "as an example, instance, or illustration." Any embodiment or implementation of the subject matter described herein as "exemplary" is not necessarily to be construed as more preferred or advantageous than other embodiments.

[0015] While this disclosure is readily adaptable to various modifications and alternatives, specific embodiments thereof have been illustrated by way of example in the accompanying drawings and will be described in detail below. However, it should be understood that this disclosure is not intended to limit it to the specific forms disclosed; rather, it is intended to cover all modifications, equivalents, and alternatives falling within the scope of this disclosure.

[0016] The term "comprises" or any other variation thereof is intended to cover non-exclusive inclusion, such that a setup, apparatus, or method that includes a list of components or steps includes not only those components or steps, but may also include other components or steps not expressly listed or inherent to such setup, apparatus, or method. In other words, in a system or apparatus, one or more elements beginning with "comprises..." do not exclude the presence of other elements or additional elements in the system or apparatus, unless further constraints are imposed.

[0017] This disclosure provides a system and method for assisting users in meeting industrial needs by dynamically routing queries to a dedicated machine learning (ML) module. The techniques of this disclosure generate adaptive workflows and utilize context-aware memory for efficient execution.

[0018] Specifically, this disclosure describes a co-pilot infrastructure framework, a high-level platform / system for facilitating seamless integration and communication of multiple Large Oracle Models (LLMs), AI / ML models, and other first-principles models in the Industrial Internet of Things (IoT) domain. The framework includes a master LLM module (referred to herein as the master agent), which dynamically determines which models or modules to invoke based on user prompts and effectively acts as the brain of the system. This module can be replaced by a custom-trained SLM or LLM to customize the decision-making process. The system supports dynamic plug-in capabilities for AI / ML models, LLMs, and other modules, ensuring scalability and flexibility. The system also features robust master configuration logic that controls the entire setup, enabling rapid customization and the introduction of new requirements and use cases. The system also includes secure and efficient connectivity to SQL databases, Blob repositories, and Cosmos DB, and integrates with Azure OpenAI PaaS and Azure Cognitive Services.

[0019] The master agent receives user queries / requests and analyzes them to determine the necessary tasks to resolve them. The master agent selects appropriate ML modules based on historical usage patterns, query context, and keyword mapping. It constructs a dynamic workflow, executing the identified modules sequentially or in parallel, and aggregating intermediate results generated by the ML modules into a final response. Additionally, the system employs a context-aware retry mechanism that adjusts the workflow based on real-time execution results, modifies the parameters used to select the ML module in case of errors, and selects an alternative module if necessary. The system also retains metadata related to user interactions, allowing for context-aware memory utilization and improved query solutions in subsequent tasks.

[0020] Figure 1An exemplary environment 100 for assisting user 110 in processing industrial needs, according to an embodiment of this disclosure, is illustrated. Environment 100 includes a system 102 deployed within an industrial environment 100. System 102 includes a master agent 104 and multiple ML modules 106. System 102 can be communicatively coupled to various industrial databases, such as databases 108a to 108n. These databases can be used to store various information related to the industrial environment 100. Master agent 104 can receive user queries / requirements. Upon receipt, master agent 104 analyzes the user query to determine the necessary tasks required to resolve the user query. Master agent 104 selects an appropriate ML module from the multiple ML modules 106 based on historical usage patterns, query context, and keyword mapping. Master agent 104 constructs a dynamic workflow to execute the identified modules sequentially or in parallel according to the dynamic workflow and aggregates intermediate results generated by the ML modules 106 into a final response. Additionally, system 102 may employ a context-aware retry mechanism that adjusts the workflow based on real-time execution results. Real-time execution results include at least one of the following: one or more errors occurring during the generation of intermediate results, and one or more retries initiated by the user. System 102 can modify the parameters used to select the ML module when an error occurs, and select an alternative module if necessary. The system also retains metadata related to user interaction, which allows for context-aware memory utilization and improved query solutions in subsequent tasks. The context-aware retry mechanism may include retrying based on one or more of the following: the context of one or more errors occurring during the generation of intermediate results, one or more retries initiated by the user, session context, and user interaction.

[0021] The detailed functions of System 102 will be described in detail in subsequent paragraphs.

[0022] Figure 2 The diagram illustrates a block diagram of a system 104 according to an embodiment of the present disclosure for performing context-aware multi-module routing and dynamic workflow execution to assist users. The system includes, but is not limited to, a main agent 202, multiple ML modules 204, I / O interfaces 206, and memory 208. All elements of system 102 work together to assist users in meeting the needs associated with industrial environment 100.

[0023] In embodiments, system 102 can be implemented on various computing platforms, such as servers, personal computers (PCs), laptops, edge computing devices, and embedded systems. In a server-based implementation, system 102 can run on cloud or local servers with high computing power suitable for industrial environments where multiple users can interact with the system simultaneously. The main agent 202 and ML module 204 can operate in a distributed manner by leveraging cloud-based APIs, databases, and computing resources to efficiently process user queries. In a PC or laptop implementation, system 102 can be installed locally to allow assistance to individual users such as engineers, analysts, and industrial operators. In another embodiment, system 102 can also be deployed on an edge computing device, where processing can occur closer to the data source (such as industrial machines, manufacturing plants, etc.). This can reduce latency and enable real-time decision-making, especially in mission-critical applications. In yet another embodiment, system 102 can be embedded in IoT-enabled devices and embedded systems, allowing seamless integration with industrial automation platforms, sensors, and control systems. Hybrid implementations are also possible, where the main agent 202 can run on a cloud-based server, while the ML module 204 and other components can run on edge devices or local computing infrastructure. This ensures scalability, efficiency, and real-time response processing based on specific deployment needs. In this way, system 102 can be customized based on industrial environments, computing requirements, and network connectivity limitations.

[0024] In this embodiment, the master agent 202 is an AI-driven orchestration engine that may include one or more large oracle modeling (LLM) and machine learning (ML) algorithms. The master agent 202 is trained to intelligently process user input indicating user requests and to identify one or more tasks to be performed to satisfy those requests. In this embodiment, the master agent 202 is trained using a structured model context protocol combined with dynamic cue tuning using Sentence-BERT and NLP-based intent recognition. The system is trained using a combination of pre-existing knowledge and domain-specific data to improve its ability to understand user intent. The master agent 202 continuously learns by optimizing its pattern recognition, enhancing contextual understanding, and improving response relevance. Additionally, it incorporates feedback from real-world interactions, allowing it to adapt and optimize performance over time.

[0025] The main agent 202 can select and invoke appropriate ML modules and aggregate their results to generate a meaningful response. The LLM is the primary inference unit, enabling the main agent 202 to understand natural language queries, infer user intent, and identify the tasks required to resolve the user query. The LLM can be supplemented by ML-based decision models that analyze historical usage patterns, task dependencies, and execution success rates to optimally select (multiple) ML modules. In embodiments, the main agent 202 can be implemented in conjunction with a dedicated hardware processing unit (not shown in the figures).

[0026] According to an embodiment, ML module 204 is also referred to as a sub-agent. Each ML module in ML module 204 is a specialized AI-driven component designed to handle specific tasks within system 102. These ML modules 204 can operate independently, but are orchestrated by master agent 202, which dynamically selects and invokes them based on user requests and identified tasks. Each ML module in ML module 204 is trained to perform a specific function and can be implemented using various AI / ML models, fine-tuned for specific industrial applications. For example, ML module 204 may include, but is not limited to, SQL modules, visualization modules, chart agents, summarization modules, API modules, etc. An SQL module can generate and execute structured queries to retrieve relevant information from an industrial database. A chart agent can process numerical data and generate visual representations such as graphs and charts. A summarization module can use NLP-based models to convert lengthy text information into concise and meaningful summaries. An API module can interact with external services to retrieve additional data or trigger automated processes. In an embodiment, the output of one module can serve as the input of another module. For example, the data retrieval module can extract machine performance data, which can then be processed by the anomaly detection module to identify deviations, followed by a visualization module generating graphical representations of the detected anomalies. Other modules can run in parallel (i.e., independently), and their outputs can be aggregated by the main agent 202. The main agent 202 and the ML module 204 have reinforcement learning-based optimization layers, which allow for continuous learning based on feedback to improve performance over time.

[0027] In an embodiment, system 102 may receive user input indicating one or more user requests via I / O interface 206. The received input may be stored in memory 208. The input may be in any form, such as, but not limited to, prompts, questions, text files, or combinations thereof. In an exemplary embodiment, memory 208 may also store instructions executable by a processing unit. The processing unit may include at least one data processor for executing program components for performing user- or system-generated requests. Memory 208 may be communicatively coupled to the processing unit or a master agent. Memory 208 may store instructions executable by the processing unit that, when executed, enable the processing unit to control various operations of system 102.

[0028] Furthermore, the main agent 202 can receive and analyze input to identify one or more tasks required to fulfill user requests. The main agent 202 can analyze historical usage patterns and task context, and also perform keyword mapping. Based on historical usage patterns, task context, and keyword mapping, the main agent 202 can identify one or more appropriate ML modules from among multiple ML modules 204 to execute the identified tasks.

[0029] Specifically, upon receiving user input, the main agent 202 first uses Natural Language Processing (NLP) techniques (such as entity recognition and semantic understanding) to analyze the user input to extract key elements. The main agent 202 can determine whether the input involves tasks such as data retrieval, analysis, visualization, summarization, or external API interaction, but is not limited to these. If the input is complex, the main agent 202 can divide it into multiple sub-tasks to ensure that each aspect of the request is executed effectively. For example, for an input such as "provide a list of assets that generated alerts in the past 24 hours and generate a summary with charts," the main agent 202 can identify four different tasks: retrieving asset alert data, aggregating the number of alerts for each asset, summarizing the results, and generating visualizations.

[0030] Once the tasks required to meet the requirements are identified, the main agent 202 can employ keyword mapping and domain-specific routing by comparing the extracted query keywords with a knowledge graph that associates specific terms with relevant ML modules. The required information for routing, such as domain information and the knowledge graph associated with each ML module, can be stored in memory 208. This process ensures that the most suitable ML module is assigned to each task's completion. For example, in an alert reporting query, the keyword "alert" maps to the SQL module responsible for data retrieval, "summary" maps to the summary module, and "chart" maps to the chart agent used for visualization. This mapping prevents misassignment of tasks and ensures that only the most relevant ML modules are selected, improving accuracy and execution efficiency.

[0031] In another embodiment, the master agent 202 can leverage historical usage patterns to determine the most reliable and efficient ML modules for execution. By analyzing past queries, module performance metrics, and execution success rates, the master agent 202 can select ML modules 204 that demonstrate higher accuracy and efficiency in similar scenarios. If multiple modules are available for a specific task, the master agent 202 can select modules that have historically performed well based on response time, accuracy, and failure rate. Additionally, the master agent 202 can evaluate the task context, which may involve analyzing dependencies and execution constraints between tasks. Some tasks may require sequential execution, while others can run in parallel. For example, the SQL module must retrieve asset alert data before the summary and visualization modules can process it, but other tasks can be executed concurrently to optimize response time.

[0032] After selecting the appropriate ML module 204, the master agent 202 can generate a dynamic workflow that defines the execution order of the selected modules and specifies input parameters for each module. The workflow is not static but designed to adjust in real-time based on the nature of the query, dependencies between tasks, and outputs. The workflow execution order ensures that tasks are executed in the correct order, passing intermediate results to subsequent modules when necessary. Some tasks may require sequential execution, with the output of one ML module serving as input to another, while other tasks can be executed in parallel to optimize processing time. The master agent can determine whether to call modules sequentially or concurrently based on dependencies and resource availability. For example, when generating an alert report, the system can first use the SQL module to retrieve asset alert data from the database and then pass it to the analysis module for data aggregation. However, once the raw data is available, the summary module and chart agent can run in parallel to reduce response time.

[0033] During execution, the master agent 202 can continuously monitor the performance / results of each ML module to ensure the successful completion of its respective task. If any module encounters an error, fails to produce the expected output, or experiences a delay, the master agent 202 can trigger a context-aware retry mechanism to intelligently resolve the issue. This retry mechanism can operate in several ways. In one embodiment, the master agent 202 can modify execution parameters to improve the chances of successful processing, such as adjusting query constraints or reformatting input data. In another embodiment, if a particular ML module repeatedly fails, the master agent 202 can select an alternative ML module with similar capabilities to handle the task. The selection of an alternative ML module can be based on historical performance data, which ensures that the master agent 202 prioritizes modules that have previously succeeded in similar contexts. In yet another embodiment, if the error is due to missing or ambiguous information in the user query, the master agent can request clarification from the user. In this embodiment, the context-aware retry mechanism includes performing retries based on one or more of the following: the context of one or more errors occurring when generating intermediate results, one or more retries performed by the user, the session context, and user interaction.

[0034] Furthermore, as the workflow progresses, the main agent 202 can aggregate intermediate results from multiple ML modules to generate the final output. The main agent 202 can combine structured and unstructured data and optimize the response format to match user preferences. Then, before being delivered to the user, the final response can be optimized through additional AI-driven processing such as summarization, formatting, and error correction.

[0035] In this embodiment, memory 208 also enables master agent 202 to retain, retrieve, and leverage past interactions to improve task execution and enhance session continuity. Master agent 202 can be configured to generate metadata corresponding to various aspects of user input processing, including user input, intermediate results, and final responses. This metadata is structured in a way that preserves the contextual flow of interactions, allowing the system to recall past queries, optimize responses based on historical data, and provide a more personalized and coherent session experience.

[0036] When the master agent 202 executes a dynamic workflow by invoking ML modules, intermediate results are generated at various stages such as data retrieval, analysis, summarization, and visualization. Memory 208 can capture and store these intermediate results along with associated metadata to ensure the maintenance of a record of input processing. Additionally, the final results can be linked to inputs and intermediate outputs to create a structured, stored representation of the entire execution. This metadata can include task dependencies, ML module selection history, execution parameters, processing errors (if any), and retry attempts to ensure a comprehensive historical record is maintained for future reference. The stored metadata allows the system to facilitate the retention and utilization of session context for subsequent processing. If a user submits a subsequent query related to a previously executed task, the master agent 202 can retrieve the stored context and provide a response based on the previous interaction. For example, if a user initially requests alert reports for a specific asset and subsequently requests trend analysis of those alerts, the system does not need to repeat the entire query execution; instead, it can retrieve the previously stored results from memory 208 and generate insights more efficiently. In embodiments, intermediate results along with associated metadata can be stored in different databases.

[0037] Furthermore, the master agent 202 can be configured to store information related to dynamic workflows in memory 208, which implements version control and reuse for context-dependent session interactions. Each dynamic workflow execution can log its task sequence, selected ML modules, execution results, and retry history, allowing the system to identify and reuse previously successful workflows for similar queries in the future. This also reduces computational overhead, improves response time, and ensures consistency in task execution. If a similar query is submitted, the master agent 202 can reference historical workflows and either fully reuse them or dynamically modify them based on real-time execution conditions.

[0038] Therefore, by using memory 208 for workflow storage and version control, the system can achieve continuous learning, improve efficiency, and enhance session intelligence. This allows the master agent 202 to provide personalized responses, intelligently handle subsequent queries, and optimize workflow execution over time, making the system more adaptive and capable of delivering highly relevant, context-aware, and effective AI-driven interactions.

[0039] Furthermore, memory 208 plays a crucial role in enhancing the context-aware retry mechanism by retaining metadata about past queries, intermediate results, final responses, and execution history. When the main agent 202 encounters errors, incomplete data, or unexpected results during workflow execution, it can intelligently modify the retry method using the stored context information, rather than blindly re-executing failed tasks.

[0040] In this way, the disclosed AI-driven orchestration method allows the system to intelligently adapt to various types of queries, dynamically optimize task execution, and provide context-sensitive and efficient responses. Unlike traditional static workflow systems that follow predefined execution paths, the master agent 202 can continuously evaluate execution conditions and make real-time adjustments, resulting in a more robust, fault-tolerant, and intelligent AI-driven system.

[0041] Figure 3 The illustration shows a system 102 according to an embodiment of this disclosure, in which a main agent 302 dynamically routes user queries to dedicated ML-based sub-agents for processing. System 102 begins with user input, received by the main agent 302. The main agent 302 may also receive a configured master prompt, which serves as an instruction to determine which module or operation should be used to resolve the user query. The master prompt guides the main agent 302 to select an appropriate method, whether by directing the request to a specific agent, retrieving relevant information, or performing necessary actions. This ensures accurate and effective query solutions.

[0042] When analyzing input, the main agent 302 identifies relevant tasks and selects appropriate ML-based sub-agents to execute them. These sub-agents may include, but are not limited to, Doc agent 306, SQL agent 308, API agent 310, graph agent 312, and agent++ 314. Doc agent 306 can be responsible for retrieving information from indexed document sources. SQL agent 308 can convert input into structured database commands and retrieve results from APM database 320. API agent 310 can enable interaction with external API 322 to integrate real-time data. Additionally, graph agent 312 can generate graphical or statistical representations. Agent++ 314 can handle specialized or undefined operations that are not suitable for predefined modules. The main agent 302 can dynamically assign tasks to these sub-agents based on the nature of the query and the requested data source. Once the sub-agents have executed their respective tasks, they return intermediate results. Intermediate results can be combined to generate a final summary response 316. The main agent 302 can continuously monitor execution and ensure data aggregation, optimization, and context enhancement are performed before generating the final response. The final response can be further structured and presented to the user in a preferred format.

[0043] Furthermore, to maintain context awareness and improve future interactions, the session history can be updated by storing relevant metadata such as inputs, intermediate results, and final responses. This enables seamless session recall and optimizes future query resolution. In this way, the system ensures intelligent task execution, efficient query routing, and dynamically adaptive responses, thereby enhancing user experience and operational efficiency.

[0044] Figure 4An exemplary flowchart illustrating the execution of context-aware multi-module routing and dynamic workflows for user assistance in an industrial environment, according to some embodiments of this disclosure, is shown. Figure 4 As shown, method 400 may include one or more steps. Method 400 may be described in the general context of computer-executable instructions. Typically, computer-executable instructions may include routines, programs, objects, components, data structures, procedures, modules, and functions that perform a specific function or implement a specific abstract data type. Method 400 may be... Figures 1 to 3 The system 102 shown in the figure is used to execute this.

[0045] The order in which method 400 is described is not intended to be construed as limiting, and any number of the described method blocks can be combined in any order to implement the method. Additionally, individual blocks can be removed from the method without departing from the scope of the subject matter herein. Furthermore, the method can be implemented in any suitable hardware, software, firmware, or a combination thereof.

[0046] In step 402, method 400 receives input including one or more user requirements related to an industrial environment. In step 404, method 400 describes analyzing the input to identify one or more tasks for satisfying the one or more user requirements. In step 406, method 400 identifies one or more ML modules suitable for performing the one or more tasks from a plurality of machine learning (ML) modules. Identification can be performed based on at least one of the following: historical usage patterns, the context of one or more tasks, and one or more keywords associated with one or more tasks. Specifically, master agent 202 can determine historical usage patterns, the context of one or more tasks, and one or more keywords associated with one or more tasks. The one or more keywords and the context of one or more tasks can be mapped to domain-specific information and historical usage patterns of the plurality of ML modules to identify one or more ML modules. The domain-specific information of the plurality of ML modules is stored in memory 208.

[0047] In step 408, method 400 generates a dynamic workflow for invoking a sequence of one or more identified ML modules. Information related to the dynamic workflow can be stored in memory 208 to allow versioning and reuse to facilitate context-sensitive session interactions. The dynamic workflow can be adjusted using a context-aware retry mechanism based on the real-time results of its execution.

[0048] In step 410, method 400 executes a dynamic workflow by invoking one or more identified ML modules according to a sequence. Each of the identified one or more ML modules can be selected to perform at least one of one or more tasks and generate intermediate results. The intermediate results generated by at least one of the identified one or more ML modules can be provided as input to at least one other ML module among the identified one or more ML modules according to a sequence. In another embodiment, the real-time results of the dynamic workflow execution can be determined. The real-time results can include at least one of the following: one or more errors in generating intermediate results and one or more retries performed by the user. The dynamic workflow can be adjusted in real time based on the real-time results using a context-aware retry mechanism. The context-aware retry mechanism includes performing retries based on the context of one or more errors in generating intermediate results, one or more retries performed by the user, session context, and user interaction. Furthermore, adjusting the dynamic workflow can include performing at least one of the following: modifying one or more parameters of one or more ML modules, and selecting (multiple) alternative ML modules for customized retries using session information and real-time result information. In step 412, method 400 generates a final result based on the aggregation of intermediate results from the identified one or more ML modules. The final result can be provided as a response to the input.

[0049] The method also describes the generation of metadata corresponding to the combination of inputs, intermediate results, and final results. The metadata indicates the session context of the inputs, intermediate results, and final results. The generated metadata is stored in memory to facilitate the retention and utilization of the session context for subsequent processing.

[0050] Unless otherwise expressly stated, the terms “an embodiment,” “an embodiment,” “multiple embodiments,” “the embodiment,” “the multiple embodiments,” “one or more embodiments,” “some embodiments,” and “an embodiment” mean “one or more (but not all) embodiments of the invention.”

[0051] Unless otherwise expressly stated, the terms “including / comprising,” “having,” and their variations mean “including but not limited to.”

[0052] Unless otherwise expressly stated, the enumerated list of items does not imply that any or all items are mutually exclusive. Unless otherwise expressly stated, the terms "a," "an," and "the" mean "one or more."

[0053] The description of embodiments having multiple components that communicate with each other does not imply that all such components are necessary. Rather, a variety of optional components are described to illustrate a wide range of possible embodiments of the invention.

[0054] When a single device or item is described herein, it will be apparent that more than one device / item (whether or not they cooperate) may be used in place of the single device / item. Similarly, when more than one device or item is described herein (whether or not they cooperate), a single device / item may be used in place of the more than one device / item, or the number of devices or programs shown may be replaced by a different number of devices / items. The functionality and / or features of a device may alternatively be embodied by one or more other devices that are not explicitly described as having such functionality / features. Therefore, other embodiments of the invention do not necessarily need to include the device itself.

[0055] Figure 4 The illustrations show certain events occurring in a specific order. In alternative embodiments, some operations may be performed, modified, or removed in a different order. Furthermore, steps can be added to the above logic, and the embodiments will still be consistent. Further, the operations described herein may be performed sequentially, or some operations may be processed in parallel. Even further, the operations may be performed by a single processing unit or distributed processing units.

[0056] Finally, the language used in this specification has been chosen primarily for readability and guidance purposes, and may not have been intentionally chosen to define or limit the subject matter of the invention. Therefore, the scope of the invention is intended to be limited not by this detailed description, but by any of the claims based on this application. Thus, the disclosure of embodiments of the invention is intended to illustrate the scope of the invention as set forth in the appended claims, and not to limit it.

[0057] While various aspects and embodiments have been disclosed herein, other aspects and embodiments will be apparent to those skilled in the art. The various aspects and embodiments disclosed herein are for illustrative purposes and are not intended to be limiting, wherein the true scope is indicated by the following claims. List of Components in the Attachment 100 Environment 102 System 104 Main Agent 106 ML modules 108a Database 108n database 110 users 202 Main Agent 204 ML module 206 I / O interfaces 208 Memory 302 Main Agent 304 main message 306 Doc Agent 308 SQL Agent 310 API Proxy 312 Chart Agent 314 Proxy++ 316 Final Summary Response 318 Index Source 320 APM Database 322 External API 324 Session History 326 users 400 methods Steps 402-412

Claims

1. A system for assisting users in performing context-aware multi-module routing and dynamic workflow execution in an industrial environment, comprising: Main agent; Multiple machine learning (ML) modules are communicatively coupled to the main agent. The primary agent is configured as follows: Receive input including one or more user requests related to the industrial environment; Analyze the input to identify one or more tasks to satisfy the requirements of the one or more users; One or more ML modules suitable for performing the one or more tasks are identified from the plurality of ML modules based on at least one of the following: historical usage patterns, the context of the one or more tasks, and one or more keywords associated with the one or more tasks; Generate a dynamic workflow for invoking the identified one or more ML modules; The dynamic workflow is executed by invoking the identified one or more ML modules according to the sequence, wherein each of the identified one or more ML modules is selected to perform at least one of the one or more tasks and generate intermediate results; as well as Based on the aggregation of intermediate results generated by the one or more ML modules through the execution of the dynamic workflow, a final result is generated, wherein the final result is provided as a response to the input to assist the user, and The dynamic workflow is adjustable based on the real-time results of its execution.

2. The system of claim 1, wherein the intermediate results generated by at least one of the identified one or more ML modules are provided as input to at least one other ML module among the identified one or more ML modules according to the sequence.

3. The system according to claim 1, further comprising: A memory, communicatively coupled to the main agent and configured to store domain-specific information of the plurality of ML modules. In order to identify the one or more ML modules, the main agent is configured as follows: Determine the historical usage pattern, the context of the one or more tasks, and the one or more keywords associated with the one or more tasks; as well as The context of the one or more keywords and the one or more tasks is mapped to the domain-specific information and historical usage patterns of the multiple ML modules to identify the one or more ML modules.

4. The system of claim 1, wherein the master agent is further configured to: Determine the real-time result of the execution of the dynamic workflow, wherein the real-time result includes at least one of the following: one or more errors that occurred when the intermediate result was generated and one or more retries performed by the user; and The dynamic workflow is adjusted in real time based on the real-time results using a context-aware retry mechanism, wherein the context-aware retry mechanism includes performing retries based on: the context of the one or more errors when the intermediate results are generated, the one or more retries performed by the user, the session context, and user interaction. In order to adjust the dynamic workflow, the master agent is configured to perform at least one of the following: modify one or more parameters of the one or more ML modules, and use session information and real-time result information to select one or more alternative ML modules for customized retries.

5. The system of claim 1, wherein the master agent is further configured to: Generate metadata corresponding to the combination of the input, the intermediate results, and the final result, wherein the metadata indicates the session context of the input, the intermediate results, and the final result; and The generated metadata is stored in memory to preserve and utilize the session context for subsequent processing.

6. The system of claim 1, wherein the master agent is further configured to store information related to the dynamic workflow in a memory to allow version control and reuse for facilitating context-dependent session interactions.

7. The system of claim 1, wherein the master agent comprises at least one of the following: one or more large language LLM models and one or more AI / ML models, and wherein the master agent is trained to select and orchestrate the plurality of ML models.

8. A method for assisting users in context-aware multi-module routing and dynamic workflow execution in an industrial environment, the method comprising: The main agent receives input including one or more user requests related to the industrial environment; The main agent analyzes the input to identify one or more tasks to satisfy the requirements of the one or more users; The master agent identifies one or more ML modules suitable for performing the one or more tasks from a plurality of machine learning ML modules based on at least one of the following: historical usage patterns, the context of the one or more tasks, and one or more keywords associated with the one or more tasks; The main agent generates a dynamic workflow for invoking the identified one or more ML modules in sequence; The main agent executes the dynamic workflow by invoking the identified one or more ML modules according to the sequence, wherein each of the identified one or more ML modules is selected to perform at least one of the one or more tasks and generate intermediate results; as well as The main agent generates a final result based on the aggregation of intermediate results from the identified one or more ML modules, wherein the final result is provided as a response to the input to assist the user. The dynamic workflow is adjustable based on the real-time results of its execution.

9. The method of claim 8, wherein the intermediate result generated by at least one of the identified one or more ML modules is provided as input to at least one other ML module among the identified one or more ML modules according to the sequence.

10. The method of claim 8, wherein identifying the one or more ML modules comprises: Determine the historical usage pattern, the context of the one or more tasks, and the one or more keywords associated with the one or more tasks; as well as The context of the one or more keywords and the one or more tasks is mapped to the domain-specific information and historical usage patterns of the multiple ML modules to identify the one or more ML modules. The domain-specific information of the plurality of ML modules is stored in memory.

11. The method of claim 8, further comprising: Determine the real-time result of the execution of the dynamic workflow, wherein the real-time result includes at least one of the following: one or more errors in generating the intermediate result and one or more retries performed by the user; and The dynamic workflow is adjusted in real time based on the real-time results using a context-aware retry mechanism, wherein the context-aware retry mechanism includes performing retries based on: the context of the one or more errors when the intermediate results are generated, the one or more retries performed by the user, the session context, and user interaction. Adjusting the dynamic workflow includes performing at least one of the following: modifying one or more parameters of the one or more ML modules, and using session information and real-time result information to select one or more alternative ML modules for customized retries.

12. The method of claim 8, further comprising: Generate metadata corresponding to the combination of the input, the intermediate results, and the final result, wherein the metadata indicates the session context of the input, the intermediate results, and the final result; as well as The generated metadata is stored in memory to preserve and utilize the session context for subsequent processing.

13. The method of claim 8, further comprising: Information related to the dynamic workflow is stored in memory to allow for version control and reuse, in order to facilitate context-sensitive session interactions.

14. The method of claim 8, wherein the master agent comprises at least one of: one or more large language LLM models and one or more AI / ML models, and wherein the master agent is trained to select and orchestrate the plurality of ML models.

Images