ALFRD

The problem

ALFRD's backend was mid-migration when I stepped in: an existing Fastify codebase with the scars of a language/framework transition, expected to support multiple organizations and real-time data feeds it wasn't originally designed for.

Constraints

• The migration was already in progress, so the fix had to work inside existing conventions and half-finished patterns, not a clean rewrite.
• Multi-tenant org management had to be bulletproof: one organization's data or session state leaking into another was not an acceptable failure mode.
• Clients needed live updates, and polling wasn't good enough, but the codebase had no existing streaming infrastructure to build on.

Architecture decisions

• 01Rebuilt multi-tenant org management around explicit tenant scoping at the data-access layer, rather than trusting every route handler to remember to filter by org.
• 02Added Server-Sent Events for live updates instead of reaching for a heavier WebSocket layer. SSE fit the one-way, server-to-client update pattern the product actually needed.
• 03Debugged my way through someone else's half-finished migration by instrumenting the boring way first: logs and reproduction, not guesswork, before touching a line of the tenant-scoping logic.

What I'd do differently

In hindsight, I'd have pushed harder up front to finish stabilizing the migration before layering multi-tenancy on top of it. Some early bugs traced back to code that was in an in-between state.