Helios

The brief

Helios began as an internal tool fighting a familiar problem: dashboards that were technically correct and practically unreadable. Teams drowned in panels, alerts, and colour. The mandate was to make a high-throughput telemetry product feel quiet — to let an on-call engineer glance once and know.

What I built

I led the platform end to end: a Rust ingestion tier writing to ClickHouse, a streaming query layer over WebSockets, and a static, near-zero-JS front end that hydrates only the live chart islands.

• Ingestion — a back-pressured Rust pipeline batching into columnar storage, holding sub-second freshness at four billion events a day.
• Query — pre-aggregated rollups plus a thin streaming API so panels update without polling.
• Interface — a restrained design system: one accent, hairline grids, and motion reserved for state changes that matter.

Outcome

p95 query latency dropped below 120ms while the surface area on screen shrank by a third. The redesign became the template for every internal product that followed.