How TReNDS Automates Root-Cause Analysis with Amazon Bedrock

The TReNDS Center at Georgia State University has built an innovative AI system leveraging Amazon Bedrock and the open-source Strands Agents SDK to automate the investigation of production errors. This system integrates AWS services such as CloudWatch, Lambda, SNS, and EKS, creating a pipeline that detects errors in real time, enriches logs with contextual data, and performs autonomous root-cause analysis.

Traditionally, engineers spent 15 to 30 minutes manually analyzing errors by examining logs and source code. The new system automates this process, enabling analysis in under 60 seconds. When an error occurs, CloudWatch filters trigger a Lambda function, which invokes the Strands Agent powered by Amazon Bedrock. The agent fetches relevant log context, retrieves source code from GitHub, and reasons through the root cause, producing a structured report sent to the team via SNS.

The architecture allows the agent to decide which tools to invoke, such as fetching source code or logs, based on error analysis. This flexibility enables deep investigation without hardcoded rules, mimicking an experienced engineer’s reasoning. The system maintains data residency within AWS, complying with health data regulations like HIPAA.

The implementation uses Amazon Bedrock models, with the Claude Sonnet model selected for its reasoning capabilities. The system’s deployment involves defining custom tools, such as source code retrieval, and orchestrating the analysis process through Lambda functions and the Strands SDK.

Since deployment, the system has significantly reduced error investigation time and provided engineers with actionable insights and suggested fixes, streamlining incident response. Future enhancements include integrating internal knowledge bases, tiered model strategies for different error types, and automated GitHub issue creation, further automating the resolution process.