Realtime analytics dashboard
2025 · product team of 6Problem
Operators waited up to a day for reports, so they acted on stale numbers and kept a parallel spreadsheet nobody trusted.
What I built
A streaming pipeline (Postgres → queue → websocket) with a virtualised table that renders 50k rows without stutter, plus saved views per team.
How it shipped
Weekly releases behind feature flags, old reports kept alive until the new numbers matched for two weeks.
Result: report latency dropped from ~24h to under 4s; the spreadsheet was retired in a month and two other teams adopted the saved views.
