AWS has launched KnowledgeForge, a solution designed to automate knowledge management in IT Service Management (ITSM). The tool extracts valuable insights from resolved incident tickets and converts them into structured knowledge base articles. This addresses the common issue of knowledge being locked in ticket history, where it remains inaccessible to future support engineers. KnowledgeForge also curates existing knowledge bases by eliminating duplicates, improving quality, and ensuring consistency. The system uses Amazon Bedrock for content generation and Amazon S3 Vectors for duplicate detection, creating a closed-loop knowledge lifecycle that enhances efficiency and accuracy in IT support operations.
KnowledgeForge operates as two interdependent subsystems: generation and curation. The generation subsystem processes clusters of resolved tickets, using Amazon Bedrock's Anthropic Claude Sonnet 4.5 model to create new draft articles and root cause analysis documents. These documents are structured with specific fields, including symptoms, root causes, and resolution steps. The curation subsystem then classifies, deduplicates, scores, and improves the content through a workflow involving AWS Step Functions and Lambda. This ensures that the final articles meet quality standards before being reviewed by a knowledge manager in ServiceNow. The system also embeds each article into Amazon S3 Vectors, enabling the generation subsystem to reuse these vectors for consistency and accuracy in future runs.
The solution is built on AWS services, including Amazon S3 for data storage, Amazon Bedrock for AI generation, and AWS Step Functions for orchestration. It requires an AWS account with access to Amazon Bedrock and specific models, as well as permissions to create and manage resources such as S3 buckets, Lambda functions, and DynamoDB tables. The code is available in the aws-samples/sample-knowledgeforge repository, and developers need familiarity with AWS CDK and Bedrock for deployment. The system is designed for large-scale document processing and offers reusable patterns for similar use cases.
Source: awsml
