TReNDS automates root-cause analysis with Amazon Bedrock
Amazon/AWS
Anthropic
Amazon Web Services
TReNDS, a neuroimaging research center, built an automated system using Amazon Bedrock and the Strands Agents SDK to investigate production errors. It detects errors in CloudWatch logs, fetches context and source code, and delivers root-cause analysis to the team via SNS, cutting investigation time from 15–30 minutes to under 60 seconds.
TReNDS (Center for Translational Research in Neuroimaging and Data Science) uses AWS since 2019, with applications on Amazon EKS and logs in CloudWatch via FluentBit. They built an architecture that automates root-cause analysis: CloudWatch subscription filters trigger a Lambda function on error patterns, the Lambda runs a Strands Agent powered by Amazon Bedrock (using Anthropic Claude Sonnet as primary model), which fetches additional log context and source code from GitHub, reasons through the error, and publishes structured analysis to SNS for email and Slack delivery. The system reduced investigation time from 15-30 minutes to under 60 seconds, with minimal Bedrock inference costs (2-3 tool-use rounds per error). They tested Claude Sonnet, Haiku, Opus, and Amazon Nova Pro/Lite, selecting Sonnet for accuracy. Deduplication via DynamoDB prevents repeated analyses for similar errors.
- Abbreviations
- AWS = Amazon Web Services — Amazon Web Services
- EKS = Elastic Kubernetes Service — Elastic Kubernetes Service
- API = Application Programming Interface — программный интерфейс приложения
- SNS = Simple Notification Service — простая служба уведомлений
- FM = Foundation Model — фундаментальная модель
- IAM = Identity and Access Management — управление идентификацией и доступом
- HIPAA = Health Insurance Portability and Accountability Act — закон о переносимости и подотчетности медицинского страхования
- ECS = Elastic Container Service — Elastic Container Service
- EC2 = Elastic Compute Cloud — эластичное облако вычислений
Source: AWS ML blog —
original