Building a Realtime Analytics Dashboard

The problem

An ops team was monitoring system health through a static report that pulled data every hour. By the time an anomaly appeared in the report, the incident had already escalated. They needed live data, not snapshots.

Scoping it down

The first request was a full analytics platform. I pushed back and scoped it to the three metrics the team actually acted on — request rate, error rate, and p95 latency. Everything else could come in a later iteration.

Getting that agreement early kept the project from ballooning.

Building it

The dashboard connected to an existing WebSocket endpoint the backend team already had running. I built the frontend in React with Recharts for the graphs — chosen for its composability over heavier alternatives like D3.

The main technical challenge was managing rolling time windows without causing re-render thrash. I ended up using a circular buffer approach to keep the last N data points in state without triggering unnecessary updates on unchanged series.

The outcome

The team went from hourly snapshots to a live view with sub-second updates. Incident response time dropped by roughly half in the first month — not because the dashboard was clever, but because the team could see problems as they happened instead of an hour later.

The scoped approach also paid off: the three-metric MVP shipped in 5 weeks, and the team used it for two months before requesting any additions.