A frontend architecture that scales with the team
A platform ran its products on copies of the same legacy frontend. We designed one architecture, wrote it down as a spec, and moved every route to it without downtime.
This is an illustrative example of how we work, not a record of a client engagement or its results.
- with module boundaries and recorded decisions
- One codebase
- cut over behind a parity check
- Route by route
Next.js
React
TypeScript
NestJS
Playwright
Argo CD
Context
The products had been forked from one frontend years earlier. Each copy had its own fixes, its own build and its own release calendar.
Challenge
A bug fix had to be written and tested once per copy, teams blocked each other, and nobody could say which rules each copy implemented.
Approach
- Wrote the spec pack first: a route inventory, the shared domain, business rules as acceptance criteria, a threat model and architecture decision records.
- Split the frontend into modules with clear boundaries: server components and streaming for content, client islands only for live data.
- Put a BFF in front of the legacy APIs, so old and new routes read the same contracts.
- Moved traffic route by route behind feature flags, each switch gated by end-to-end tests that compare the old and new routes.
What changed
Every product now ships from one codebase. Teams work in their own modules in parallel, and a new product is a spec change reviewed like any other.