Fathom

Fathom’s ops team ran the business out of a dozen spreadsheets. We designed a single source of truth they could act on, not just admire.

Ops dashboards are where good intentions go to die — too dense, too slow to change. This one had to stay useful as the business shifted under it.

• Dense information that stays scannable
• Alerts that mean something
• A layout that adapts to new metrics
• Trust — numbers that always reconcile

We shadowed the team for a week, learned which numbers actually drove decisions, and designed around those.

• A priority-led information hierarchy
• Thresholds and alerts co-designed with ops
• A flexible grid for metrics yet to exist
• A component kit their engineers could ship

A Go collector streams service heartbeats over WebSockets into TimescaleDB; the React/TypeScript console renders incident state machines, and Grafana handles the long-tail dashboards ops already trusts.

• Go collector: 40k heartbeats/s on one small node
• Incident lifecycle as an explicit state machine
• WebSocket fan-out with backpressure handling
• Timescale continuous aggregates for uptime bars

Also considered. Node for the collector — Go's goroutines and memory profile won the ingestion path.