Case Study / 03 — Data Platform

Pulse
Analytics

Two million events per second is not a dashboard problem, it's a physics problem. Pulse renders live infrastructure telemetry in WebGL — dashboards that feel less like reports and more like flight instruments.

Role
UX, Rendering Architecture, Frontend
Timeline
2025 — 6 months
Stack
WebGL, WebSockets, ClickHouse, Rust
Outcome
2M events/s at 60fps · MTTR −44%

The
firehose

The existing tool aggregated everything into one-minute buckets — incidents were visible a minute after they mattered. Engineers had learned to distrust their own dashboards and tail logs instead.

We moved aggregation to the edge and rendering to the GPU. The browser receives a delta stream over WebSockets and draws it as instanced geometry — DOM charts topped out at ~3,000 points; the WebGL pipeline holds 1.2 million on screen without dropping a frame.

  • Rust edge aggregators emit 60hz delta frames, not raw events.
  • Every chart is one draw call — instanced quads with data in texture memory.
  • Anomalies pulse and bloom before thresholds fire: preattentive, not paged.

The
instrument

The design language borrows from avionics: dark surfaces, luminous data, zero chrome between the operator and the signal. Color is reserved exclusively for state — if something glows, it needs a human.

During its first on-call quarter, mean time to resolution dropped 44%. The unexpected metric: engineers started leaving Pulse open on a spare monitor because it was calming to watch. That's what observability should feel like.

0Events rendered
0Sustained
0Faster resolution
0Draw call per chart
Next case Nova Bank ⟶