AWS's KnowledgeForge turns ITSM tickets into curated KB articles

KnowledgeForge, an AWS sample solution, tackles a classic enterprise IT problem: resolved ITSM tickets hold valuable fixes but their knowledge stays trapped in ticket history, while knowledge bases grow bloated with duplicate and stale articles. KnowledgeForge works both sides of that gap. It mines clusters of resolved incident tickets to auto-generate draft knowledge base articles and root cause analyses, then curates every article—both new and existing—through four steps: classifying by type, detecting duplicates, scoring quality, and rewriting weak content.

The architecture uses Amazon Bedrock for content generation and improvement, Amazon S3 Vectors for efficient duplicate detection via embeddings, and AWS Step Functions to orchestrate the curation pipeline on AWS Lambda. Generation runs on Amazon ECS with AWS Fargate. The system closes the loop by embedding every article into an Amazon S3 Vectors index, which generation reuses as grounding on subsequent runs—so new articles avoid duplicating existing knowledge. Articles flow into ServiceNow for knowledge-manager review and approval, with decisions written back to DynamoDB. The solution requires access to Claude Sonnet 4.5 and Titan Text Embeddings V2, and uses CDK for deployment.