Appen ARG Dashboard

Appen (NASDAQ: APT) is a $500M+ AI training data company whose pipelines supply Google, Microsoft, and Amazon. When they needed internal tooling to manage 3 separate client performance programs simultaneously, we built the infrastructure from scratch—without access to production data during development.

The technical challenge was multi-tenancy done right: TCS, ByteWorks, and DataPower have completely different data structures, KPI definitions, and reporting cadences. Prisma + PostgreSQL with row-level isolation keeps client data strictly separated while a shared Next.js 16 monorepo (5 apps, 10 packages) means UI changes propagate everywhere. Upstash QStash handles the async report generation job queue—CSV ingestion triggers a pipeline that runs analysis, generates PDF reports, and dispatches email—all without blocking the UI.

The real innovation was demo mode. Enterprise B2B sales die when you can't show the product without exposing real client data. We engineered a production-grade mock layer using MSW (Mock Service Worker) + tRPC interceptors that operate at the request level—not fake static screens. Four strategies: MOCK_FOREVER with seeded data and periodic heartbeat updates, SKELETON_FOREVER for loading state showcases, ERROR_FOREVER for testing error recovery UX, and MIXED for high-fidelity module-specific simulation. A unified 'schema validate → redaction → normalize → value protocol' pipeline ensures every mock response passes the same validation as real data. Role switching (Viewer/Analyst/Admin) and live data seed control available in the demo panel.

This is the demo mode architecture that every enterprise SaaS eventually builds—we shipped it on day one.