Cadence

Cadence needed an analytics platform that its own team could be proud of — fast, honest with numbers, and light enough that non-analysts would actually open it.

We took it from empty repo to live product: the data model, the interface, the charts, the query layer and the polish.

Analytics tools drown people in options. The hard part was not drawing charts — it was deciding what not to show, and making the few things that mattered instant.

• Sub-second queries over large datasets
• A chart system that stays legible when data is ugly
• Dashboards a beginner can build in minutes
• A visual language that reads as trustworthy

We designed the information architecture first, then built vertically — one real dashboard, wired to real data, before we scaled the pattern across the product.

• Typed query layer with aggressive caching
• A restrained chart kit built on a single grid
• Server-driven layouts, no bespoke pages
• Motion used only to explain change, never to decorate

Event ingestion lands in Kafka, compacts into ClickHouse for sub-second aggregate queries, and a Next.js app on Vercel streams server-rendered dashboards; heavy rollups run as scheduled materialised views.

• Kafka topics per event class, replayable backfills
• Row-level security tokens minted per workspace
• Edge-cached public share links with signed URLs

Also considered. TimescaleDB + Redis was the simpler path — ClickHouse won on cardinality and cost per billion rows.