How TReNDS automates root-cause analysis with Amazon Bedrock

This article describes how TReNDS Center built an automated root-cause analysis system using Amazon Bedrock and the Strands Agents SDK to investigate production errors in real time.

• CloudWatch subscription filters detect error patterns and invoke Lambda functions for analysis
• Strands Agent with Claude Sonnet model autonomously investigates errors using custom tools
• Custom tools fetch source code from GitHub and surrounding log context from CloudWatch
• Agent produces structured analysis with severity, root cause, code context, and suggested fixes
• Investigation time reduced from 15-30 minutes to under 60 seconds per error
• Data residency maintained within AWS account, supporting HIPAA compliance requirements
• Results published to SNS for email and Slack notifications to engineering teams
• Deduplication using DynamoDB prevents duplicate analyses for repeated errors

The agentic approach enables flexible, adaptive error investigation without hardcoded decision trees, delivering production-ready root-cause analysis and suggested fixes automatically to engineering teams.