How tripleten-com/ai-system-engineering-showcase works
TripleTen is a production-shaped incident-response showcase. An engineer triggers a deterministic incident in the War Room; the API generates telemetry and evidence, sanitizes logs, retrieves a runbook, and runs a checkpointed agent plan that pauses for human approval. Approval dispatches a validated job to a worker, which records simulated remediation, archives a postmortem, and reports completion so telemetry can recover and the UI can display the outcome. The graph preserves the documented storage and observability integrations. Source excerpts are partial, so uncertain internal call wiring is omitted.
Main entry points: consumer.py Job Consumer · graph.py Agent Graph · api.ts API Client · orchestrator.py Run Orchestrator
Components 29
Incident War Room
- War Room UI
services/incident-war-room/src/App.tsx - Incident Stream
services/incident-war-room/src/hooks/useIncidentStream.ts - SSE Client
services/incident-war-room/src/services/sseClient.ts - API Client
services/incident-war-room/src/services/api.ts - RAG Inspector
services/incident-war-room/src/components/RagInspectorView.tsx
API Control Plane
- FastAPI App
services/incident-agent-api/src/incident_agent_api/main.py - Incident Routes
services/incident-agent-api/src/incident_agent_api/api/routes/incidents.py - Stream Routes
services/incident-agent-api/src/incident_agent_api/api/routes/stream.py - Retrieval Routes
services/incident-agent-api/src/incident_agent_api/api/routes/retrieval.py
Analysis and Planning
- Run Orchestrator
services/incident-agent-api/src/incident_agent_api/agent/orchestrator.py - Agent Graph
services/incident-agent-api/src/incident_agent_api/agent/graph.py - Guardrails
services/incident-agent-api/src/incident_agent_api/agent/guardrails.py - Evidence Sanitizer
services/incident-agent-api/src/incident_agent_api/security/sanitizer.py - Hybrid Search
services/incident-agent-api/src/incident_agent_api/retrieval/hybrid_search.py - Runbook Ingestion
services/incident-agent-api/src/incident_agent_api/seed/ingest.py
Telemetry and Evidence
- Telemetry Engine
services/incident-agent-api/src/incident_agent_api/telemetry/engine.py
Async Remediation
- Worker Daemon
services/remediation-worker/src/remediation_worker/main.py - Job Consumer
services/remediation-worker/src/remediation_worker/consumer.py - Remediation Handlers
services/remediation-worker/src/remediation_worker/handlers - Postmortem Archive
services/remediation-worker/src/remediation_worker/postmortem.py - Completion Callback
services/remediation-worker/src/remediation_worker/callback.py
Other components
- Engineer
- Shared Contracts
packages/contracts/src/tripleten_contracts - PostgreSQL pgvector
- Redis
- SQS Queues
- S3 Storage
- Prometheus
- Jaeger
Connections 32
- Engineer → War Room UI: triggers / approves
- War Room UI → API Client: requests actions
- API Client → Incident Routes: HTTP calls
- API Client → Retrieval Routes: searches runbooks
- Incident Routes → Run Orchestrator: starts / authorizes
- Run Orchestrator → Telemetry Engine: transitions run state
- Run Orchestrator → Agent Graph: invokes workflow
- Agent Graph → Evidence Sanitizer: sanitizes evidence
- Agent Graph → Hybrid Search: retrieves guidance
- Agent Graph → Guardrails: validates tools
- Run Orchestrator → SQS Queues: dispatches approved job
- Agent Graph → PostgreSQL pgvector: checkpoints state
- Runbook Ingestion → PostgreSQL pgvector: stores runbooks
- Hybrid Search → PostgreSQL pgvector: searches runbooks
- Telemetry Engine → Prometheus: publishes metrics
- Telemetry Engine → Stream Routes: publishes frames
- Stream Routes → SSE Client: streams events
- Incident Stream → SSE Client: subscribes
- Incident Stream → API Client: fetches snapshot
- War Room UI → Incident Stream: renders state
- RAG Inspector → API Client: probes retrieval
- Worker Daemon → Job Consumer: polls queue
- Job Consumer → SQS Queues: receives jobs
- Job Consumer → Remediation Handlers: executes approved tools
- Job Consumer → Redis: checks idempotency
- Job Consumer → Postmortem Archive: records results
- Postmortem Archive → S3 Storage: archives report
- Job Consumer → Completion Callback: reports completion
- Completion Callback → Incident Routes: notifies API
- Worker Daemon → Redis: refreshes heartbeat
- FastAPI App → Jaeger: exports traces
- Shared Contracts → War Room UI: defines payload types