This project shows how AI-powered multi-agent systems can revolutionize cloud incident response. Built using Amazon Bedrock (Claude & Nova), OpenAI embeddings, and advanced monitoring with AWS CloudWatch and CloudTrail, LangSmith used for debugging this solution detects, diagnoses, and resolves incidents—eliminating manual JIRA ticket creation and dramatically reducing response times.
-
Continuous AWS Monitoring:
Detects issues across data pipelines, compute resources, storage, and access boundaries in real time. -
RAG-Powered Diagnosis:
Finds solutions using a vector store of previous incidents; escalates to web search for novel problems. -
Autonomous Ticket Creation:
Generates detailed JIRA tickets with incident context, root cause, and actionable remediation steps. Later, sends a Push Notification to Team. -
Multi-Agent Architecture:
- Supervisor Agent: Orchestrates the workflow and manages agent handoff state.
- Monitoring Agent: Watches AWS logs and resources, detects anomalies.
- Diagnosis Agent: Uses RAG (vector DB + Tavily web search) for recommended fixes.
- Resolution Agent: Creates JIRA tickets for each incident, sends a Push Notification to Team.
-
Self-Learning Memory:
The system’s knowledge base grows with every new incident and solution.
This system is built for real-world, collaborative cloud operations. At the American Red Cross, our team uses it to monitor critical AWS services, automate incident response, and reduce operational overhead:
- 24/7 Monitoring: The system continuously tracks data pipelines, compute resources, storage, and access patterns, catching incidents the moment they occur.
- Automated Triage & Context: Instead of sifting through scattered logs and alerts, engineers receive actionable JIRA tickets for every significant event—complete with root cause, context, and step-by-step remediation.
- Self-Improving Knowledge Base: Each new incident handled expands the system’s internal memory, enabling even faster, more accurate responses for future issues.
- Seamless Collaboration: Support, DevOps, and security teams are notified instantly and can trust that every ticket includes the right context and remediation plan—streamlining handoffs and reducing time-to-resolution.
- Reduced Alert Fatigue: Teams focus only on actionable, high-priority incidents with recommended fixes—helping them stay proactive, not reactive.
How to Start Using It in Your Team:
- Deploy the system to monitor the AWS resources your team relies on.
- Integrate the JIRA ticketing automation into your support or DevOps workflow.
- Let the agents handle detection, diagnosis, and documentation—your engineers focus on solutions and improvements.
- As the memory database grows, incident response becomes faster, more consistent, and more intelligent.
At the American Red Cross, this solution has transformed the way our technical teams respond to incidents.
Manual triage is now a thing of the past. We’ve cut response times, increased transparency, and enabled true collaboration across operations, support, and security.
- Python
- Amazon Bedrock (Claude, Nova)
- OpenAI Embeddings
- LangChain
- AWS CloudWatch & CloudTrail
- JIRA API
- Tavily (Web Search API)
- LangSmith