Multi-tenant operations
Users work inside isolated organizations with role-based permissions around services, incidents and operational data.
CASE STUDY · INCIDENT OPERATIONS · APPLIED AI · 2026
PULSEOPS AI
A full-stack incident operations platform combining uptime monitoring, realtime coordination, public status communication and evidence-grounded AI assistance inside one operational workflow.
VIEW SOURCE CODE ↗WHY I BUILT IT
Operational AI is useful when it works from evidence and stays inside a controlled incident process.
PulseOps AI was designed as an incident-management system first and an AI feature second. Services, monitor checks, outages, incident timelines and customer-facing status all exist independently of the model.
The AI layer consumes that operational evidence to help operators form hypotheses, prioritize next actions, prepare status communication and draft postmortems. Generated output remains advisory and is persisted for traceability rather than silently automating remediation.
INCIDENT PIPELINE
01Monitor CheckWorker tests service health
→02IncidentAffected services + timeline
→03AI AnalysisEvidence-grounded assistance
→04ResolutionStatus + postmortem
CORE SYSTEMS
Uptime monitoring
A Redis/BullMQ worker performs HTTP checks, records monitor history and uses retry/backoff before confirming outages.
Incident lifecycle
Incidents connect affected services, authored timeline updates, health state and resolution into one traceable workflow.
Realtime coordination
Redis-backed events keep operational views and public status information synchronized as incident state changes.
Evidence-grounded AI
AI analysis receives stored incident evidence and returns probable causes, confidence indicators, next actions and customer-update drafts.
Postmortems
Resolved incidents can generate AI-assisted postmortems while preserving known unknowns and keeping human review in the publication loop.
OPERATIONS FLOW
01monitor
02detect
03coordinate
04communicate
05resolve
Monitoring and incident state are linked: checks establish evidence, incidents organize response, realtime events synchronize views, public status communicates impact and resolved incidents become reviewable postmortems.
TECHNICAL STACK
Next.js · React · TypeScript · authenticated operations dashboard
Fastify · TypeScript · RBAC · SSE / operational endpoints
PostgreSQL · Prisma · persisted incidents, monitor history and AI artifacts
Redis · BullMQ · HTTP monitoring · retry/backoff
OpenAI Responses API · Docker Compose · lint · typecheck · builds
ENGINEERING DECISIONS
Worker isolation
Monitoring runs in its own BullMQ worker instead of tying recurring network checks to the web request lifecycle.
Evidence before AI
Incident analysis is grounded in stored service, timeline and monitoring evidence rather than asking the model to infer an outage from a vague prompt.
Human publication
AI can prepare customer-facing drafts, but publication and operational decisions remain controlled by human operators.
Persisted artifacts
AI outputs are stored with the operational record so generated analysis can be reviewed alongside the incident that produced it.
WHAT THIS PROVES
The product includes relational data, background workers, Redis coordination, monitoring logic and incident state beyond a standard CRUD application.
AI is constrained to evidence-backed assistance and explicit human review rather than being presented as an autonomous root-cause authority.
Internal response workflows and public status communication share the same incident source of truth.
Web, API, PostgreSQL, Redis, workers and AI form one coherent operational architecture.
NEXT CASE STUDYARABDEVKIT ↗