AI Agent Execution Plans for Cloud Network Troubleshooting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Cloud networks with thousands or millions of nodes face challenges in maintaining node and function health, leading to processing delays, increased costs, and customer dissatisfaction due to inefficient update mechanisms, inconsistent documentation quality, and high costs of large language model (LLM) usage.
Innovation Solution
AI agents generate executable plans from source documents like TSGs and MOPs during a planning phase, converting them into flowcharts with nodes and edges, and execute these plans using script and semantic tools, reducing LLM usage and hallucinations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If LLM is used to solve incidents in real-time, then troubleshooting capability is improved, but cost and hallucination risk increase
Solution Approach 1:
The patent pre-generates executable plans from documentation during a planning phase before incidents occur. These plans are stored and can be directly executed during incidents, eliminating the need to call LLMs in real-time for plan generation. This preliminary action reduces both cost and hallucination risk while maintaining troubleshooting capability.
Solution Approach 2:
The patent creates executable copies of troubleshooting documentation that can be directly executed. Instead of having the LLM interpret and generate plans during incidents, pre-generated executable representations are used, reducing LLM usage to verification only and significantly lowering cost and hallucination potential.
2Adaptability or versatility
If LLM is used to execute documentation at incident time, then adaptability is improved, but reliability decreases due to hallucinations
Solution Approach 1:
Executable plans are generated in advance during a planning phase, allowing thorough verification and validation before deployment. This preliminary generation ensures consistency and reliability while maintaining adaptability through the structured executable format that can be applied to various incidents.
Solution Approach 2:
The system incorporates verification steps where the generated executable plan is checked against the original documentation and incident context. This feedback mechanism ensures the plan is both adaptable to the specific incident and reliable in its execution, catching potential hallucinations before they cause problems.
3Loss of energy
If traditional manual troubleshooting is used, then cost is reduced, but time to mitigation increases
Solution Approach 1:
The system enables automated self-service troubleshooting by executing pre-generated plans without requiring manual engineer intervention for each step. This automation significantly reduces time to mitigation while keeping costs low by using LLMs only for plan generation and verification, not for every troubleshooting action.
Solution Approach 2:
The patent replaces the mechanical process of manual engineer execution with automated execution of executable plans. Scripts and tools are automatically run based on the generated plan, substituting human manual operations with automated systems that are both faster and more cost-effective at scale.
4Manufacturing precision
If documentation is made highly detailed and precise, then execution accuracy is improved, but document complexity increases
Solution Approach 1:
The patent uses automated plan generation and execution systems to handle the complexity of precise documentation. The system automatically parses, validates, and executes detailed procedures, freeing documentation from needing to be both highly detailed and simple. The executable format naturally handles complexity while ensuring precision in execution.
Solution Approach 2:
The patent transforms documentation from a static text format to an executable parameter-driven format. By changing the representation parameters of documentation into structured executable plans with defined inputs, outputs, and validation rules, the system achieves high execution accuracy without requiring documentation to be manually complex.
Data Source
AI summary
A computing system receives an indication of an operational procedure to be performed in the computing network. The operational procedure is represented as a structured operator-readable document comprising operator-executable operations for resolving an issue in the computing network or implementing a modification to the computing network. Content from the operational procedure is input to an artificial intelligence (AI) agent to generate a plan for executing the operational procedure in the computing network. The plan includes a plurality of operations and at least one network tool for executing the operations. The generated plan is verified to meet one or more predetermined criteria.