James Chang / Projects / Aleph / Dashboard
Aleph · Product · Last Updated
Admin Dashboard
From a question to dozens of iterations to one final prompt. This is what AI-assisted product engineering looks like.
No wireframe. No Figma file. No PRD. This dashboard started with a question — “what does an importer need to see first?” — and evolved through dozens of iterations with Claude Code: researching patterns, prototyping layouts, testing data shapes, refining the insight engine. The prompt below is the final artifact — the distilled specification that produces this dashboard from scratch. It’s the same process I use across all my projects.
claude "Build an executive compliance dashboard for Aleph"
Design process
- 12 parallel research agents analyzed architecture, performance, security, TypeScript, data integrity, UX
- Stripe/Linear "hero + tile grid" over tabbed explorers — executives decide in 5 seconds; tabs bury the signal
- Every metric has 3 layers: the number, the context (delta + sparkline), the AI insight
- Okabe-Ito colorblind-safe palette
- Rule-based insight engine (zero cost) + Claude API enrichment (weekly)
Layout
- Hero: "Compliance Coverage" — 81% big number, progress bar, inline AI insight
- 6 Stat Tiles (3-col grid): big number + delta + sparkline + AI insight + drill-through
- Est. MRR: $290 · Users: 48 · Activation: 24%
- Products: 106 · Alerts: 298 · Doc Risk: 16
- 3 Funnels: Onboarding (-76% drop) · Compliance · Retention
Key decisions
- "Hero + tiles" not tabs — executives decide in 5 seconds
- Inline insights — context right where you need it
- Rule-based + AI hybrid — 80% value at zero cost
- Coalescing cache — 871ms cold, <10ms warm
- tabular-nums everywhere — no layout shift
- Gray-first, color for meaning — Okabe-Ito for colorblind safety
Stack
- React 18 + TypeScript + Vite + shadcn/ui
- Tailwind + Recharts (sparklines,
isAnimationActive=false) - TanStack Query (
staleTime: 5min) - Express + PostgreSQL + Drizzle ORM
- InsightEngine (6 rules) + AIInsightService (Claude, weekly)