• 

  logo

• 

  logs

• 

  logs

Inspiration

Ops/support teams struggle with fragmented signals—support tickets, error logs, uptime pings—without a unified incident view. Fireline aggregates these signals into incidents, auto-triages them, and generates communications.

What it does

Fireline is a real-time incident room that:

• Ingests heterogeneous support events via Redpanda (Kafka-compatible streaming)
  
• Aggregates events into incidents grouped by service and environment
  
• Uses LLM-based triage to assign severity (SEV1/SEV2/SEV3), owner, and summaries
  
• Generates internal and external status updates
  
• Sends Slack notifications to assigned engineers and escalates SEV2+ incidents
  
• Provides a real-time web UI showing incidents and status updates

How we built it

• Backend: Java 21 + Akka SDK (Agents, Consumers, KeyValueEntities, Views, HTTP Endpoints)
  
• Streaming: Redpanda for event ingestion
  
• LLM: OpenAI API for triage and communications generation
  
• Frontend: Next.js + TypeScript for the incident room UI
  
• Observability: Distributed tracing with Jaeger (OpenTelemetry)
  
• Notifications: Slack integration for real-time alerts

Challenges we ran into

• Tracing setup: Switched from Akka Console to Jaeger for reliable trace visualization
  
• Prompt injection: Implemented input validation and structured JSON parsing to mitigate LLM prompt injection
  
• Distributed testing: Used Akka SDK TestKit patterns for testing event-driven components
  
• Component orchestration: Coordinated Agents, Entities, and Consumers via ComponentClient

Accomplishments that we're proud of

• End-to-end pipeline: Events → incidents → triage → comms → UI working end-to-end
  
• Distributed tracing: Full visibility into the incident processing pipeline with Jaeger
  
• Comprehensive testing: Unit, integration, and agent tests with 90%+ coverage
  
• Slack integration: Real-time notifications with team-based routing and escalation
  
• Production-ready architecture: Durable state with KeyValueEntities, proper error handling, and observability

What we learned

• Akka SDK patterns: Agents, KeyValueEntities, Views, and component communication
  
• Distributed tracing: OpenTelemetry integration and trace visualization with Jaeger
  
• LLM safety: Prompt injection mitigation and structured output validation
  
• Event-driven architecture: Building resilient systems with Kafka-compatible streaming

What's next for Fireline

• More event types: Integrate monitoring systems (Datadog, New Relic) and chat platforms
  
• Richer UI: Real-time updates via WebSockets, incident timeline visualization, and status page integration
  
• Production hardening: Rate limiting, authentication, and multi-tenant support
  
• Advanced triage: ML-based incident correlation and automated remediation suggestions

Built With

• akka-sdk-3.5.9
• docker
• jaeger
• java-21
• junit
• maven
• next.js-16
• openai-gpt-4o-mini
• python
• react-19
• redpanda
• slack-api
• tailwind-css-4
• typescript-5