As the company scaled its data pipelines and internal systems, diagnosing issues became increasingly complex and time-consuming.

Teams had to manually:

• Check logs, databases, and queues across multiple systems
• Investigate failures without a centralized diagnostic tool
• Piece together insights from fragmented sources

This led to:

• Slow root cause analysis
• Limited visibility into system health
• Reactive rather than proactive issue detection
• High dependency on engineering time for investigations

A fully deployed AI-powered Investigator Agent, designed to automate diagnostics, reporting, and system monitoring.

The system:

• Aggregates data from multiple sources (databases, logs, queues)
• Performs automated root cause analysis using an AI agent (ReAct-based reasoning)
• Enriches investigations with external and internal knowledge sources
• Generates structured reports after each run
• Sends real-time Slack alerts, including:
  
  • Error notifications
  • Heartbeat signals confirming system activity
  • Summary insights with links to full reports

The solution operates through scheduled runs and alert-based triggers, ensuring continuous monitoring without manual intervention.

• Full visibility into system health and pipeline status
• Automated investigation and reporting after each run
• Significant reduction in manual diagnostic work
• Faster root cause identification across complex systems
• Increased confidence through continuous monitoring and heartbeat alerts

Delivered as a working MVP Investigator system, fully deployed in the client environment.

Key implementation components:

• AI agent built with LangGraph (ReAct pattern) and LLM reasoning
• Backend services using Node.js with direct PostgreSQL querying
• AWS infrastructure (CloudWatch, S3) for logs, traces, and report storage
• Integrations with internal tools and external knowledge sources
• Slack-based communication layer for alerts and reporting

Team:
1 Backend/AI Engineer
1 DevOps
1 Tech Lead
1 PM
2 QA

Timeline:

• POC (diagnostic scenarios): ~3 weeks
• Full MVP with Slack integration: ~5 weeks