The TReNDS Center at Georgia State University, in collaboration with Georgia Institute of Technology and Emory University, has developed an architecture that leverages Amazon Bedrock to automate root-cause analysis. This system integrates Amazon CloudWatch, AWS Lambda, and the Strands Agents SDK to provide real-time error detection and structured analysis. The solution aims to streamline incident response by reducing the manual effort required to investigate errors. The architecture is designed to keep data within the same AWS environment, ensuring compliance with data residency requirements for health-related research data. Source: awsml
The TReNDS team faced challenges with manual error investigation, which often took 15–30 minutes for straightforward issues and longer for complex problems. They recognized the potential of using a foundation model to automate this process, allowing the model to analyze logs, retrieve source code, and produce structured root-cause analysis without hardcoding the investigation path. This approach enables the system to dynamically adapt to different error scenarios by using the Strands Agents SDK to orchestrate tool usage. The model reasons through errors by pulling log context and source code, providing a more efficient and scalable solution for incident response. Source: awsml
The architecture involves applications running on Amazon EKS sending logs to CloudWatch via FluentBit. CloudWatch subscription filters trigger Lambda functions upon detecting error patterns, which then invoke the Strands Agent powered by Amazon Bedrock. The agent fetches source code from GitHub, analyzes logs, and publishes a structured analysis to an SNS topic for team delivery. This setup ensures that all data remains within the AWS environment, meeting HIPAA compliance requirements for health-related research. The solution is applicable to various AWS services and on-premises workloads that send logs to CloudWatch. Source: awsml
