AUTONOMOUS TASK FLOW AND LLM-BASED OPERATION AUTOMATION METHOD AND SYSTEM

Abstract

This invention relates to an autonomous task flow and LLM-based operation automation method and system (1). The invention includes the following steps: the event chain and process flow monitoring module (6) extracts process flows from execution records in target systems (5); the process and entity dependency model module (15) provides a structured dependency model; the error and deviation detection module (7) identifies anomalies; the entity flow validation unit (18) detects entity flow errors by referencing expected transitions; the intelligent task plan generator (8) generates the appropriate task plan in the form of a directed graph; the agent manifest generator (16) derives the machine-readable agent manifest (19); the agent runtime (12) executes the agents within the framework of this manifest; the phased deployment module (17) implements automatic rollback on metric decline; and the monitoring, validation, and feedback module (13) enables closed-loop learning by passing the results back.

Description

1 TARIFF AUTONOMOUS TASK FLOW AND LLM-BASED OPERATIONAL AUTOMATION METHOD AND SYSTEM This invention relates to execution records, request-response records, event records, transaction traces, or 5 In software systems that generate inter-service communication data, including but not limited to, IT operational systems, business process automation platforms, technical support and incident in management systems, data processing and reporting systems, and large language model supported systems. Developed for use in digital assistant applications, for one or more clients (3) autonomous task flow used between one or more servers (2) to which it is connected and 10 It is related to the LLM-based operation automation method and system (1). In traditional software-based systems, task steps, operational workflows, and Solution plans are usually prepared by developers, operators, or system administrators. They are defined manually. In existing automation solutions, the rules are pre-defined. The workflows to be executed are being determined, maintained as fixed templates, and new process types are being introduced. Additional rules, scripts, or configurations need to be written for this. In these configurations, the main language... The model's use is often limited to chatting, accessing information, or generating help text. remaining, task learning from system behavior, inferring solution flows from similar events, and Dynamic task generation cannot be implemented. Specifically, process flow in distributed systems. Because it spread across multiple services, event chains, call traces, and messaging mechanisms, error 20 or the resolution of deviation situations is often the responsibility of human operators, system administrators, and Autonomous software agents are becoming dependent on user intervention. The current situation... methodology, work guide automation, process automation, and digital assistant solutions, It generally fails to derive executable task plans from past system behavior, new In these situations, they cannot automatically generate and implement these plans and, according to the task results, 25 It does not offer an agent infrastructure that continuously improves itself. Furthermore, the current systems are... which assets, tokens, or structured outputs are carried over to the next step of the process Because it cannot semantically verify that the information has progressed through transmission, error detection is often... This is done solely based on error codes, timeouts, or sequence disruptions. Therefore, adaptation to process changes remains limited, requiring manual steps for new task types. The need for modeling arises, and the operational burden falls on the user. One or used between one or more servers (2) to which multiple clients (3) are connected With autonomous task flow and LLM-based operation automation method and system (1), the target 2 Execution logs, event chains, application interactions, request-response generated in systems (5) Automatic extraction of transaction flows from sequences and similar usage data, this Detecting errors, interruptions, deviations, and anomalies in the flows is necessary for task execution. The process and entity dependency model consists of endpoint, entity type, and dependency information. The use of module (15) allows for the creation of task plans appropriate to the identified situation. automatic generation in this form, input from large language model agents from these task plans- By deriving output schemes, permission limits and role rules, agent manifestos (19) the production of major language model agents within the framework of the aforementioned manifesto constraints structuring and execution, measuring task outcomes, phased deployment and controlled distribution through automatic recall mechanisms and the resulting 10 The use of results in subsequent task production, standardization, and process modifications. Automatic adaptation, independent task execution by large language model agents, and faster problem solving. The solution involves writing manual workflows, defining manual work guides, and repetitive operational tasks. By reducing interventions, learning from logs and event streams, and providing feedback to itself. updating, performing error detection based on asset flow verification, agent behavior 15 Automatically derives from a structured task and dependency model and is controlled incremental. providing deployment mechanisms and a secure operational automation infrastructure. that is intended. Autonomous task flow and LLM-based operation to achieve the invention's purpose. The automation method and system (1) are shaped in the attached figures, of which; 20 Figure 1. Autonomous task flow and LLM-based operation automation method and system. The parts in the figures are numbered, and their corresponding parts are shown below: 1. Autonomous task flow and LLM-based operation automation method and system. 2. One or more servers 3. One or more clients 4. Applications 5. Target systems 30 6. Event chain and process flow monitoring module 7. Error and deviation detection module 8. Smart task plan generator 3 9. Job description repository 10. Task execution engine 11. Agent design module 12. Agent working time 13. Monitoring, verification and feedback module 5 14. Network communication units 15. Process and Entity Dependency Model Module 16. Agent manifesto producer 17th stage commissioning module 18. Asset flow verification unit 10 19th Agent Manifesto Autonomous task flow and LLM-based operation automation method and system (1), a or multiple servers (2), one or more clients (3), applications (4), target systems (5), event chain and process flow monitoring module (6), error and deviation detection module (7), smart 15 task plan generator (8), task description repository (9), task execution engine (10), agent design module (11), agent runtime (12), monitoring, verification and feedback module (13), process and entity dependency model module (15), agent manifesto (19) producing agent manifesto manufacturer (16), phased commissioning module (17), asset flow verification unit (18) and It consists of network communication units (14) that provide data communication between modules. 20 Autonomous task flow and LLM-based operation automation method and system (1), a or one or more clients (3) are connected via network communication units (14) It runs on multiple servers (2). Autonomous task flow and LLM-based operation. Automation method and system (1), applications with multiple software components (4) 25 includes and each software component is on a different server or multiple servers (2) It is able to work. Autonomous task flow and LLM-based operational automation method and Since the system (1) consists of multiple microservices, each implementing a different service, applications which are software components (4), load balancing, accessibility and scalability According to the rules, it is located on one or more servers (2). Each software has 30 Applications which are components (4) exchange data with other software components, Depending on the process performed, one or more clients (3) and one or more servers (2) Their roles can be reversed. Furthermore, autonomous task flow and LLM-based operational automation are possible. method and system (1), with third-party software systems, adapters, connector interfaces, 4 Integration occurs through plugins, network hook components, or intermediary services, facilitating communication. It can be found. Autonomous task flow and LLM-based operational automation method and in the system (1), applications (4), client applications, services or external that trigger the system The actors can be the target systems (5), which are databases, application programming interfaces. units or microservice infrastructures, messaging systems, event handling components and 5 It can include similar distributed software environments. Process and entity dependency model module (15), task generation and task execution (5) endpoint, service, process step, entity type in the target systems, to be used during It is a unit that provides a structured model representing data dependencies and transition relationships. 10 The model in question derives its design from pre-defined system definitions, at runtime. from observations, service discovery logs, configuration metadata, or any of these It can be obtained from the combination. The process and entity dependency model module (15), this Using this information, we can determine which step needs which type of input, and which step requires which type of input. that it produces the output and that the transitions between steps in a specific task flow are based on which data or 15 It represents that this is possible with the types of entities involved. Thus, task plan generation and The necessary semantic framework for asset flow validation is provided. In the autonomous task flow and LLM-based operation automation method and system (1), Event chain and process flow monitoring module (6), application on target systems (5) 20 interactions, execution logs, request-response sequences, service calls, events triggers, message queue flows, trace IDs, correlation IDs, and time By tracking their signatures, it extracts transaction flows. Event chain and transaction flow tracking. module (6) brings together temporally and logically related records to form a specific reconstructing the process event chain, sequencing the flow steps, and analyzing the system behavior. It identifies normal patterns. Thus, by moving beyond individual records, it analyzes processes. The actual execution sequence, dependency structure, and task step patterns are obtained. Event with the chain and process flow monitoring module (6) temporally and logically related events chains, correlation IDs, track IDs, session ties, time proximity, transaction order, recurring parameter values, cross-service call dependencies, and 30-minute intervals between events Cause-and-effect relationships are derived using patterns that show the chain of events and processes. With the flow monitoring module (6), event chain inference rule-based matching, pattern-based correlation, statistical association, or one or more of the learning methods This can be done using the Event Chain and Process Flow Monitoring Module (6), in addition repetitive process sequences, paths to successful completion, flows that end in failure, It identifies the steps that create bottlenecks and the frequently repeated intervention patterns. Event chain and process flow monitoring module (6), by grouping records belonging to the same process, process-based flow It can create maps, task step sets, and solution patterns. Event chain and Process flow monitoring module (6), contextual data for next task plan production phase 5 It provides. Error and deviation detection module (7), event chain and process flow monitoring module (6) abnormalities, errors, interruptions, delays, sequence disruptions, and missing elements in the streams generated by the system. The step is repeated failed attempts, unexpected responses, or out-of-standard deviations. It determines that there is no error or deviation. The error and deviation detection module (7) detects the expected flow patterns. comparing actual workflows, error code density, processing time deviation, sequence violation, missing elements By using indicators such as event, repetition rate, failure rate, and so on, anomalies are detected. The error and deviation detection module (7) uses statistical threshold methods, if necessary. deviation analysis, pattern matching techniques, or machine learning-based classifiers 15 It can be used. The error and deviation detection module (7) detects only that there is a problem. not only, but also the affected process step, the relevant service or endpoint, and the specific location of the deviation. It can determine how the disease developed at each stage and identify potential root cause candidates. Thus, the following steps can be taken... The task production phase requires not only raw error information but also contextually enriched information. Event data is transmitted. Error and deviation detection module (7) with error or deviation confidence 20 score, number of observed anomalies, failure repetition, processing time deviation, sequence violation, missing the number of steps involved, the number of affected services or components, the level of consistency with past similar events, and The solution is determined by a weighted assessment of similarity to the past. Error and With the deviation detection module (7), low confidence deviations are detected only in monitoring or recommendation mode. It can be maintained, and high-security deviations enable automated task plan generation. It can trigger it. Asset flow verification unit (18), process and entity dependency model module (15) By referencing the dependencies defined by, a task flow or observed At each step of the process chain, the asset types expected in the output of the previous step are 30 that the tokens or structured fields are correctly passed to the next step, It checks that it has not been transferred. Asset flow verification unit (18), at each step transition the existence of expected assets, type accuracy, formal consistency, value integrity, and It checks the validity status. The absence of an asset token in its expected location indicates a type of 6 It contains incompatibility, empty or invalid values, or has expired. In this case, an asset flow error is generated. The asset flow verification unit (18) detects entity stream errors to the intelligent task plan generator (8) and task execution engine (10) This information can be transferred to a new agent node to address the missing asset transfer. adding, restructuring an existing step, choosing an alternative path, or 5 It can be used to re-execute a specific step. This allows for error detection. It is not limited to just error codes or timeouts, but also includes the semantics of the task flow. Verification at the level becomes possible. The intelligent task plan generator (8) detects errors and deviations. in accordance with the situation determined by the detection module (7) and asset flow verification unit (18) It is the unit that creates the solution plan or workflow. The intelligent task plan generator (8) has been in the past 10 observed successful solution flows, task templates, process patterns, policies rules, target system (5) capabilities and process and entity dependency model module (15) It generates executable task plans using dependency information. Smart task During the production of the task plan by the plan producer (8), the type of incident identified, affected service, process context, available tools, authorizations, prerequisites and success criteria are taken into consideration. 15 The intelligent task plan generator (8) generates task plans in the form of a directed graph. It produces. In this graph structure, each node handles an API call, an agent operation, and a It can represent a verification step or a decision point, typographical between nodes. Edges, on the other hand, represent the type of entity, or data, that is transferred from the output of one node to the input of another node. It shows the diagram or control flow dependency. In this way, task plans are arranged in a straight line. Instead of step lists, it can also include parallel branching, conditional progression, and merge points. They are expressed as structured task diagrams. With the intelligent task plan generator (8), Task plan extraction, sequential pattern mining, probabilistic transition modeling, similarity measurement, Graph alignment is achieved using embedding-based methods or combinations thereof. This can be accomplished. Thus, the extracted patterns include not only the call order, but also the same 25 Semantics includes which types of entities are transferred between which steps in time. Wealth can be transformed into job descriptions. The intelligent task plan generator (8) is a pre-learned task in the form of an application. By matching the existing event context with the established patterns, we can select the appropriate task plan, another 30 In terms of implementation, a new task workflow is used with planning supported by a large language model. It is able to synthesize the generated task plan, triggering conditions, prerequisites, sequential and parallel operations. steps, conditional branching rules, tools to be used, parameter diagrams, entity flow expectations, approval requirements, timeout rules, retry policies, feedback 7 This can include areas such as acquisition steps and success criteria. Thus, the task plan not just a natural language description, but a machine-executable, structured task description. It is transformed into. The intelligent task plan generator (8), during task plan selection or synthesis, event context, past task success rates, tool accessibility, process risk level, user The need for intervention, policy rules, and job security score together 5. Candidate task plans are being evaluated based on historical success rate, frequency of repetition, and implementation. They can be ranked according to criteria such as cost or priority, and are suitable for automated execution. Plans that require user approval can be distinguished from other plans. Task definition store (9), task 10 created by intelligent task plan generator (8) It is the unit where job definitions and workflow templates are stored. In the job definition repository (9), the job Definitions can be stored in YAML, JSON, or similar structured formats, for each task. Description: version number, trigger rule, prerequisite set, tool list, parameter schema, asset flow expectations, confidence level, buyback steps, and success metrics It can be versioned. The task definition store (9) also contains 15 past task versions. It ensures that the information is stored in an unchanging manner and that the evolution of the task is monitored. The task execution engine (10) uses the task definitions found in the task definition store (9). by taking these and executing them on target systems (5) in a way that is consistent with the directed graph structure It is a unit. The task execution engine (10) determines the order of the nodes in the task diagram, 20 dependencies, parallel execution branches, conditional branching rules, prerequisites, time to observe the overrun rules, retry policies, and recall procedures, as necessary. When this happens, the agent coordinates one or more agents during the working time, thus completing the task. It ensures the execution of correction, verification, on the target systems (5). Data collection, configuration update, restart, call repetition, notification sending 25 or similar operations can be implemented in a controlled manner. Task execution engine (10), By coordinating with the asset flow verification unit at each step of the task diagram, that the expected asset tokens or structured outputs have been transferred correctly It can verify this. When the expected entity type is not found in the output of a step, or the type When the incompatibility is detected, the task execution engine (10) repeats the relevant step 30 being able to try, choose an alternative path, safely stop the task, or By sending feedback to the manufacturer (8), the smart task plan will address the missing asset transfer. It can request an agent node or task correction. The task execution engine (10), repetitive execution during task execution, step-by-step status tracking, intermediate error management, and 8 It can provide policy violation checks. If necessary, for specific steps, the user Approval may be awaited, actions in certain risk classes may be restricted, or tasks may be assigned. By partially implementing the system, a safe stop can be applied. Agent design module (11) provides the tools that large language model agents will use, request 5 defining templates, capabilities, permission limits, policy restrictions, and contextual resources The Agent Design Module (11) is a module that designs agents suitable for a specific task type or process context. By creating profiles, we can determine which agents can use which tools under which conditions. The agent design module (11) determines the mapping between the mission plan and agent capabilities. By doing so, it prepares the agent configuration to be used in the execution of the mission. Agent work 10 time (12), agent design module (11) and agent manifest generator (16) are configured with Runtime when agents become active and large language model calls are processed. It is the environment. In this environment, context management, intermediate state storage, tool calling, step-by-step Reasoning, response generation, and task progress are managed. Agent work time. (12) to produce action and execute tasks using the tools defined in the task plan 15 It works in coordination with the engine (10). Agent runtime (12) of each agent The manifest includes a list of permitted vehicles or endpoints, speed limits, and protected areas. It applies its policies during working hours, as defined in the manifesto. an attempt to access a non-existent resource, exceeding the speed limit, or an operation outside the protected area. If requested, the relevant call may be rejected, and the violation record will be monitored, verified, and returned within 20 days. It can be reported to the power supply module (13). The monitoring, verification and feedback module (13) measures the outcome of the task being performed, evaluating the alignment of task outputs with expected results and providing feedback to subsequent task production. It is the module that provides the feed. Monitoring, verification and feedback module (13), task success 25 rate, solution time, error rate, number of retries, user intervention requirement, Asset flow validation success rate, improvement in target metrics, and similar performance. It monitors its indicators. Monitoring, verification and feedback module (13), live task unexpected results in implementation, policy violations, vehicle failures, or low performance. When accuracy issues are detected, the relevant task version can be deactivated, the previous 30 It can suggest or automatically trigger a rollback to the stable version. Monitoring, verification and feedback module (13), from the phased commissioning module (17) also records the results obtained and sends the smart task plan to the generator (8), task definition 9 by transferring to the repository (9), agent design module (11) and agent manifest generator (16), the future This enables the improvement of tasks. Thus, the confidence scores of successful task plans increase. plans that can be improved, fail, or require high user intervention have low security. They can be marked as such, plan preference rankings can be updated, and more information can be obtained in new events. This makes it possible to select appropriate task flows. Task plan, Agent 5 configuration or specific task steps, error rate, low confidence score, policy incompatibility, unexpected working time result or high risk class a certain threshold When its value is exceeded, it can be temporarily deactivated or put into restricted execution mode. It can be linked to user approval or reverted to the previous stable version. This increases the security of automated execution and prevents erroneous task plans from occurring in System 10. This prevents it from having uncontrolled effects. Agent manifest generator (16) is a subcomponent of the agent design module (11) or It operates as a complementary unit and from the process and entity dependency model module (15) and the smart mission plan uses information from the mission plan received from the manufacturer (8) every 15 It generates a structured agent manifest (19) for an agent node. Agent manifest (19) the types of entities and data that the agent expects to receive from previous nodes in its task diagram. the input schema including the fields, entity types expected to be provided to subsequent nodes, and data The output schema, which includes the fields, endpoints that the agent can call at runtime, or vehicle list, speed limits, quota information, protected area rules, user approval 20 a machine-readable representation that includes the required operation classes, role rules, and confidence score It is a configuration document. Input and output diagrams are derived from the typed edges in the task diagram and It is derived from domain mappings in the dependency model. Agent manifest generator (16), agent It automatically derives the manifesto (19), and in the necessary application forms, the user Approval to review, approve or amend the agent manifesto (19) 25 It can offer a mechanism. During the production of the agent manifest (19), persistent identifiers, Expected entity types, required fields, and allowed tool calls are automatically configured. Confidence levels can be assigned to each parameter or edge. Phased commissioning module (17), a newly created or updated task 30 the plan or agent manifesto (19) has direct control over all traffic or all event streams It is the unit that allows the process to be tested on a controlled subset, rather than being activated immediately. With the phased commissioning module (17), a new task plan or canary execution phase can be implemented. Agent manifest (19) first provided a configurable total traffic or event stream The percentage is being considered, and within this subset, the task's success rate, solution time, and error rate are being analyzed. The rate, asset flow verification success rate, and user intervention requirement are currently stable. It is compared with the previous version. If the success criteria remain above the acceptance thresholds, The traffic share of the version can be gradually increased. Below certain thresholds in the metrics. Performance observation: significant increase in error rate or rise in asset flow errors 5 If detected, the new version will be automatically deactivated, and the system will revert to the previous stable version. It is reversible. This rollback decision is based on configurable threshold values ​​and rule-based principles. Some applications may rely on evaluations or statistical significance tests. These formats may require user approval. In the autonomous task flow and LLM-based operation automation method and system (1), Execution logs consisting of interactions between applications (4) and target systems (5), Request-response sequences, event logs, and service calls are monitored through the event chain and process flow module. (6) collected and temporally and logically related events of a process The chains are being extracted. Simultaneously, the process and entity dependency model module (15), 15 task plan and endpoint, entity type and dependency relationships required for task execution It provides. By running the error and deviation detection module (7) on the extracted flows, the error, Interruptions, delays, sequence violations, missing steps, or similar anomalies are identified. Asset flow The validation unit (18) references the expected transitions in the dependency model at each step. checking whether the expected assets have been moved correctly and asset flow 20 It identifies their mistakes. The intelligent task plan generator (8) creates a solution plan suitable for the identified situation, past solution patterns, task templates, policy rules, existing tools, dependency model, and The task creates a workflow in the form of a directed graph using reliability indicators. 25 The generated task plan includes new agent nodes for the steps where asset flow errors were detected. It is possible to add or define alternative routes. The created plan includes a task definition. It is recorded in the repository (9). Simultaneously, the agent design module (11) and the agent The manifest generator (16) prepares the input / output schemes of the agents in accordance with the task schedule, permission In preparing the boundaries, speed limits and protected area policies, the agent's working time is 30 (12) makes the relevant agents ready to execute. The mission execution engine (10) prepares the mission. taking the task in the definition repository (9) and using the agents (12) during the agent runtime, It performs its task on target systems. 11 During execution, the asset flow verification unit (18) is expected at each step transition. In validating asset tokens and structured outputs, relevant issues are identified when problems are detected. The step can be repeated, an alternative route can be chosen, or the task can be completed safely. It can be stopped when a new mission plan or agent manifest (19) is put into effect. The phased commissioning module (17) first starts the canary execution, then the new version of the traffic 5 or trying on a configurable subset of event flows, success metrics Comparing it to the current stable version, the traffic share increases when the acceptance thresholds are exceeded. It is gradually increased, and when a metric decline or an increase in asset flow errors is detected, It automatically rolls back to the previous stable version. The result of the process is then analyzed by the monitoring, verification and feedback module (13). The process involves tracking task success status, resolution time, error and retry logs, and asset stream. validation results, phased rollout metrics, and the need for user intervention. After evaluation, the results are sent to the smart mission plan generator (8), mission definition repository (9), and agent. The design module (11) and the agent manifest are transmitted to the generator (16). Thus, the system 15 not only able to perform tasks but also to update oneself by learning from past experiences. It creates a closed-loop operation automation structure. In the autonomous task flow and LLM-based operation automation method and system (1), Automatically from (5) actual usage data and event chains in target systems 20 Task flows are being learned, and errors or deviations are identified in the form of directed graphs. Structured task plans are being created from the process and entity dependency model module. (15) agent input / output diagrams and agent manifesto (19) definitions are derived, entity stream With verification, error detection is taken to the semantic level, manifesto of large language model agents. It is operated in a controlled manner within the limits of its constraints, with phased commissioning and automatic 25 With recovery mechanisms in place, new mission plans are distributed securely, and the results obtained are... The results are used in future task production. Thus, the manual work guide writing, manual process modeling and repetitive operational interventions are reduced, process standardization, speed of solution, explainability, secure automation, and operational efficiency. is being increased. 30

Claims

12 REQUESTS 1. Autonomous task flow and LLM-based operation automation method and system (1) and its feature is; one or more servers (2), one or more clients (3), one or Applications with multiple software components on multiple servers (2) (4), 5 databases, application programming interfaces, or microservice infrastructures, messaging systems, event handling components, and similar distributed software environments including target systems (5), application interactions in target systems (5), execution records, request-response sequences, service calls, event triggers, message queues By tracking streams, track IDs, correlation IDs, and timestamps, 10 Event chain and process flow monitoring module (6) which extracts the process flows, chain and Anomalies, errors, on the flows generated by the process flow monitoring module (6), interruption, delay, sequence disruption, missing step, repeated failed attempts, Error and deviation determine whether there is an unexpected response or non-standard deviation. Detection module (7), Error and deviation detection module (7) and asset flow verification unit 15 Smart that creates a solution plan or workflow appropriate to the situation determined by (18). Task plan generator (8), task created by smart task plan generator (8) The task definition repository (9), where the definitions and workflow templates are stored, task definition by taking the job descriptions in the repository (9) and putting them on the target systems (5) Task execution engine (10) that executes in accordance with the directed graph structure, big language 20 the tools, request templates, capabilities, and permission limits that the model agents will use, Agent design module (11) that defines policy constraints and context resources, agent Active of configured agents with design module (11) and agent manifest generator (16) The agent is the runtime environment in which large language model calls are processed and the process is carried out. working time (12), measuring the outcome of the task performed, the expected 25 of the task outputs. assesses the alignment with the results and provides feedback for subsequent task generation. Monitoring, verification and feedback module (13), endpoint on target systems (5), services, process steps, entity types, data dependencies, and transition relationships are represented Process and entity dependency model module providing a structured model (15), Agent manifesto producer (16), newly created or 30 (19) all traffic or all of an updated mission plan or agent manifest (19) Instead of directly enabling it on event streams, a controlled subset The phased commissioning module (17) which enables testing on the dependency By referencing the expected transitions in the model, the expected assets are correct at each step. 13 It checks whether the assets have been transferred correctly and detects asset flow errors. The network that provides data communication between the asset flow verification unit (18) and the modules. It consists of communication units (14).

2. According to Claim 1, autonomous task flow and LLM-based operation automation method and The system (1) is characterized by its ability to detect session connections (5) on target systems, time proximity, 5 the order of operations, recurring parameter values, and inter-service call dependencies. and by using patterns that show cause-and-effect relationships between events, application interactions, execution logs, request-response sequences, service calls, events triggers, message queue flows, trace IDs, correlation IDs, and By tracking timestamps, event chain inference, rule-based matching, pattern 10 one of the following methods: correlation based on statistical association or learning methods. Using several methods, event chaining and transaction flow tracing extract transaction flows. the module (6) brings together temporally and logically related records, reconstructing the event chain of a specific process, sequencing the flow steps, and It identifies normal patterns of system behavior and records the same process. 15 by grouping, process-based flowmaps, task step sets and solutions to create patterns and contextual data for the next task plan production phase It is a verification.

3. According to Claim 1, autonomous task flow and LLM-based operation automation method and The system (1) has the feature of comparing the expected flow patterns with the actual flows, error 20 Code density, processing time deviation, queue violation, missing event, repetition count, failure ratios and similar indicators or statistical threshold methods, deviation analysis, patterns error and matching techniques using machine learning-based classifiers by the deviation detection module (7), event chain and process flow monitoring module (6) Anomalies, errors, interruptions, delays, sequence disruptions, and missing elements in the extracted streams. 25 The step is a repeated failed attempt, an unexpected response, or an out-of-standard deviation. and the affected process step, the relevant service or endpoint, and the specific location of the deviation. It indicates that it developed in the stage and identifies possible root cause candidates, low confidence. Its ability to keep deviations in only monitoring or suggestion mode provides high security. It can trigger automatic task plan generation in case of deviations. 30 4. According to Claim 1, autonomous task flow and LLM-based operation automation method and The system (1) is characterized by the process and entity dependency model module (15). 14 By referencing the defined dependencies, a task flow or observed process At each step of the chain, the types of entities expected in the output of the previous step, whether the tokens or structured fields are correctly passed to the next step the asset flow verification unit (18) which checks that it has not been transferred, at each step transition the existence of expected assets, type accuracy, formal consistency, value integrity 5 and verifying its validity, ensuring that an asset token is where it is expected to be. its absence, type incompatibility, containing empty or invalid values, or If it has timed out, it will generate an asset stream error, which it detected. entity stream errors to the intelligent task plan generator (8) and task execution engine (10) In order to transfer this information and address the missing asset transfer, a new agent node 10 addition, restructuring an existing step, choosing an alternative path or it can be used to re-execute a specific step.

5. According to Claim 1, autonomous task flow and LLM-based operation automation method and system (1) and its feature is; previously learned task patterns in an application form By matching the current event context, triggering conditions, prerequisites, sequential and parallel 15 steps, conditional branching rules, tools to be used, parameter diagrams, assets flow expectations, approval requirements, timeout rules, retry appropriate task including areas such as policies, rollback steps and success criteria error and deviation detection module (7) and asset flow verification unit (18) that can select its plan Creating a solution plan or workflow appropriate to the situation determined by, in the past 20 observed successful solution flows, task templates, process patterns, policies rules, target system (5) capabilities and process and entity dependency model Using the dependency information in module (15), each node can make an API call, an agent operation can represent a verification step or a decision point, The type of edges between nodes, from the output of one node to the input of another node, is 25 Parallel to showing the type of entity transferred, data schema, or control flow dependency. in the form of a directed graph including branching, conditional progression, and merge points The intelligent task plan generator (8) that produces executable task plans, task plan inference, sequential pattern mining, probabilistic transition modeling, similarity measurement, Graph alignment uses embedding-based methods or combinations thereof and 30 In another application form, with large language model-supported planning, the context of the event, past task success rates, tool accessibility, process risk level, user the need for intervention, policy rules and task confidence score together It is the ability to synthesize a new task flow that is being evaluated.

6. According to Claim 1, autonomous task flow and LLM-based operation automation method and The system (1) is characterized by the task generated by the intelligent task plan generator (8). The task definition repository (9), which is the unit where the definitions and workflow templates are stored, 5 Job descriptions can be stored in YAML, JSON, or similar structured formats. Each task definition includes a version number, trigger rule, prerequisite set, and tools list. parameter diagram, asset flow expectations, confidence level, retrieval steps, and success. versioning based on criteria and also the immutability of past mission versions This ensures that the data is stored in a way that allows the evolution of the task to be monitored. 10 7. According to Claim 1, autonomous task flow and LLM-based operation automation method and The system (1) is characterized by the tools that large language model agents will use, request templates, capabilities, permission limits, policy restrictions, and contextual resources defining agent profiles suitable for a specific task type or process context 15 that constitute, determine which agents can use which tools under which conditions Agents configured with the agent design module (11) and agent manifest generator (16) the runtime environment where it becomes active and large language model calls are processed The agent's working time (12), context management, intermediate state storage, tool call, Step-by-step reasoning, response generation, and task progress management. ensuring, using the tools defined in the task plan, producing action and task 20 It works in coordination with the execution engine (10) and each agent in the manifest specified list of permitted vehicles or endpoints, speed limits and protected areas whether it complies with the policies, as defined in the manifesto, during working hours. attempting to access a non-existent resource, exceeding the speed limit, or being outside a protected area. In the event of a transaction request, the relevant call may be rejected and the violation record may be monitored and verified. 25 and can be reported to the feedback module (13).

8. According to Claim 1, autonomous task flow and LLM-based operation automation method and The system (1) is characterized by measuring the result of the task performed, the task outputs. assessing the alignment with expected results and providing feedback for subsequent task production. Monitoring, verification and feedback module (13), task success rate, solution 30 duration, error rate, number of retries, user intervention requirement, presence flow validation success rate, improvement in target metrics, and similar performance. 16 monitoring indicators, unexpected results in live task execution, policy When violations, vehicle failures, or low accuracy situations are detected, the relevant It can deactivate the task release and suggest reverting to the previous stable version. or can be triggered automatically, obtained from the phased commissioning module (17) recording the results obtained and sending the intelligent task plan to the generator (8), task definition store 5 (9), by transferring to the agent design module (11) and agent manifest generator (16), the future It ensures the improvement of tasks.

9. According to Claim 1, autonomous task flow and LLM-based operation automation method and system (1) and its feature is that it is a sub-component of the agent design module (11) or It functions as a complementary unit and consists of 10 modules from the process and entity dependency model module. (15) and information obtained from the mission plan from the smart mission plan manufacturer (8) agent manifest generator (16) using persistent identifiers, expected entity Types, mandatory areas and permitted vehicle calls can be automatically determined for each By assigning a confidence level to a parameter or edge, it is automatically generated and the user can... 15 submitted with an approval mechanism for review, approval or modification a structured agent manifest (19) for each agent node, the agent’s task the types of entities and data fields it expects to receive from previous nodes in its graph the input schema containing the types of entities and data expected to be provided to subsequent nodes The output schema, which includes the fields, is the endpoints that the agent can call at runtime. or vehicle list, speed limits, quota information, protected area rules, 20 including transaction classes requiring user approval, role rules, and trust score. It is a machine-readable configuration document.

10. According to Claim 1, autonomous task flow and LLM-based operation automation method and The system is (1) and its feature is to take the job descriptions from the job description repository (9), 25 that execute these on target systems (5) in accordance with the directed graph structure. The task execution engine (10), which is a unit, determines the order of the nodes in the task diagram. dependencies, parallel execution branches, conditional branching rules, prerequisites, timeout rules, retry policies, and rollback steps monitoring, when necessary, one or more agents during agent operation time by coordinating, ensuring the execution of the task and 30 on the target systems (5) correction, verification, data collection, configuration update, restart, call This means that repetition, sending notifications, or similar actions can be implemented in a controlled manner. 17 11. According to claim 10, repetition during task execution is ineffective, step-by-step situation. Task execution that can provide monitoring, intermediate error management, and policy violation auditing. its engine (10), with the asset flow verification unit at each step transition of the task diagram by working in a coordinated manner, the expected asset tokens or structured outputs The ability to verify that it has been transferred correctly, the expected asset in the output of a step is 5. If the species is not found or a species mismatch is detected, repeat the relevant step. being able to try, being able to choose an alternative path, being able to safely stop the mission or by sending feedback to the intelligent task plan producer (8) regarding the missing asset transfer The ability to request a new agent node or mission correction that will resolve the issue is necessary. In certain situations, it may wait for user approval for specific steps, in a specific risk class 10. by limiting actions or by partially carrying out the task, a safe stop It is the ability to implement it.

12. According to Claim 1, autonomous task flow and LLM-based operation automation method and system (1) and its feature is to be used during task production and task execution, (5) endpoint, service, process step, entity type, data dependency and 15 in the target systems from pre-established system definitions representing transitional relationships, the study from time observations, service discovery logs, configuration metadata or a unit that provides a structured model derived from a combination of these. Using this information, which process and entity dependency model module (15) can determine which What type of input each step requires, what type of output each step produces 20 and the transitions between steps in a specific task workflow, which data or asset to represent what is possible with its types and task plan production and asset flow It provides the necessary semantic basis for verification.

13. According to Claim 1, autonomous task flow and LLM-based operational automation method and The system (1) has the feature that the asset flow verification unit (18) every 25 during execution. Validating the expected asset tokens and structured outputs at the step transition, When a problem is identified, the relevant step should be repeated, an alternative course of action should be chosen, or... safely halting the mission, a new mission plan or agent manifesto (19) When commissioned, the staged commissioning module (17) first performs canary execution By initiating it, the new version allows for the configuration of traffic or event streams within a configurable sub-section 30. testing it in the cluster, comparing success metrics with the current stable version, acceptance. The metric involves gradually increasing traffic share when staying above certain thresholds. 18 When a decline or increase in asset flow errors is detected, an automatic rollback is applied. It means reverting to the previous stable version.

14. According to Claim 1, autonomous task flow and LLM-based operation automation method and The system (1) has the feature of; monitoring, verification and feedback module of the process result. (13) Analysis by the task success status, solution time, error and re-analysis. test logs, asset flow verification results, phased deployment metrics And by assessing the need for user intervention, the results are given to the intelligent task plan producer. (8), to the mission definition repository (9), to the agent design module (11) and to the agent manifest generator (16) transmission, the system not only performs the task but also transmits past executions A closed-loop operation automation that can update itself by learning. 10 It is the creation of its structure.

15. Autonomous task flow and LLM-based operation automation method (1) and its feature is; - Execution records, request-response sequences and events in the target systems (5) The chains are collected by the event chain and process flow monitoring module (6) and processed Extraction of flows, 15 - The process and entity dependency model module (15) endpoints (5) in the target systems, providing a structured model representing entity types and dependency relationships -Anomalies of the error and deviation detection module (7) on the extracted flows determination and asset flow verification unit (18) process and asset dependency model Detecting entity flow errors by referencing expected transitions in module (15) 20 to do, -The intelligent task plan is a directed diagram of the task plan that is appropriate to the identified situation by the manufacturer (8). to create in this way, -From the agent manifest producer's (16) process and entity dependency model module (15) and Input-output diagrams, permission limits, and policy 25 for each agent node from the mission plan. derivation of the machine-readable agent manifesto (19) containing the rules, - Agent working time (12) agents to carry out within the framework of manifesto restrictions, -The phased commissioning module (17) controlled the new task plan or manifest. testing on the subset and implementing automatic rollback on metric decline, 19 -The results of the monitoring, verification and feedback module (13) are used in the smart task plan. to the producer (8), to the mission definition repository (9), to the agent design module (11) and to the agent manifest by transferring it to the producer (16) and enabling closed-loop learning It includes the steps. 10 20 30

Images