Releases were frozen. One week to fix it.
A multi-market commerce platform mid-takeover: pages failing to render, deployments crashing the site, unknown dependencies cascading failures — and a client openly questioning the engagement.
backend-for-frontend
THE SITUATION
The platform served 30+ markets through a micro-frontend architecture over a shared BFF. During a takeover, releases had effectively stopped: a single MFE or a mismatched dependency version could take down an entire page, and nobody could reliably say which change caused which failure. Trust between client and delivery team was at its lowest point.
WHAT I DID
- Refactored rendering with a builder pattern so one component failure could no longer take down a whole page — failures degraded gracefully instead.
- Mapped every BFF dependency and inter-MFE version so the team could finally reason about what a release actually changed.
- Sat with DevOps through the deployments themselves rather than reviewing from a distance — the fastest way to see the real failure modes.
- Wrote a short release runbook so the fix held after I stepped back.
WHAT CHANGED
The platform stabilised, releases resumed, and the engagement survived the takeover. I became the first voice on every critical bridge call after that — and the discovery work that followed set the client’s internal quality benchmark.
“Do whatever you think would fix this.”