Scope-Based IT Failure Investigation with LLM Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing IT system failure investigations are time-consuming and costly due to the need for manual review of vast amounts of configuration and observation data, and existing natural language processing technologies are not directly applicable for automating failure investigations.
Innovation Solution
A computer system utilizing a large language model (LLM) to generate findings for IT system failures by identifying relevant elements, converting observation data into interpretable text, and generating prompts for analysis, thereby automating the failure investigation process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual failure investigation is performed by operation administrators, then detailed failure analysis can be conducted, but the process is time-consuming and costly
Solution Approach 1:
The system enables self-service failure investigation by having the LLM autonomously analyze observation data, identify failure causes, and generate investigation reports without requiring operation administrators to manually review extensive logs and configuration information
Solution Approach 2:
The patent replaces the mechanical manual investigation process with an automated LLM-based system that uses natural language processing to understand failure contexts, retrieve relevant information, and generate analyses, thereby eliminating the time-consuming manual review of vast amounts of data
2Productivity
If automation programs and rules are generated for failure investigation, then investigation efficiency improves, but development costs become extremely high due to enormous number of variations
Solution Approach 1:
The system changes the approach from developing complex automation programs with numerous rules to using an LLM that can adapt to different failure scenarios through parameter changes in prompts and data input, achieving flexibility without requiring extensive programming
Solution Approach 2:
Instead of creating custom automation programs for each failure scenario, the system uses a general-purpose LLM that can be prompted to handle various failure types by copying and adapting successful analysis patterns from training data, significantly reducing development costs
3Extent of automation
If existing natural language processing technologies are used, then some language processing tasks can be automated, but they cannot be directly applied to failure investigation
Solution Approach 1:
The patent applies a universal LLM that can perform multiple functions including understanding failure contexts, retrieving relevant configuration and observation data, analyzing failure causes, and generating investigation reports, making the system adaptable to various failure scenarios without requiring scenario-specific processing
Data Source
AI summary
A computer system is coupled to an IT system formed of elements, and a text generation system. The computer system holds element information for managing observation data obtained from the IT system, relation information for managing relevance between the elements, and scope information that defines a scope representing a range of an investigation. The computer system selects the scope in a case where a failure has occurred in the IT system, identify at least one element to be investigated based on the relation information and the selected scope; obtain the observation data relating to the identified at least one element from the element information, generate a first prompt, which includes the viewpoint corresponding to the selected scope, the identified at least one element, and the obtained observation data, and which instructs to output findings relating to the failure in the IT system, and input the first prompt.