Live monitoring surface for distributed services. Charts as sentences — data that speaks before you ask it a question.
Dashboards are usually walls of numbers. The team needed a surface that surfaced anomalies before they became incidents — something you could glance at, not study.
A Rust backend ingesting service metrics over a fanout WebSocket pipeline, paired with a SvelteKit frontend that renders sparklines and plain-language summaries side by side. When a service dips, you see the chart and a sentence like "p99 latency up 40 % since 14:22."
The hardest part wasn't the data pipeline — it was deciding what not to show. Every metric added cognitive load. Cutting the panel count in half made the remaining panels actually useful.