The starting point
Meridian ran their patient scheduling on a ten-year-old Rails monolith that nobody on the team felt comfortable changing. Outages were frequent, deploys took an hour, and three of the original engineers had moved on. Every new feature request — even small ones — turned into a multi-week negotiation about whether the system could survive the change.
We were brought in to lead a replatform: same product, same workflows, modern foundations, and a team that could move again.
What we did
We spent the first two weeks not writing code. We mapped the existing system end-to-end, talked to the schedulers who actually used it every day, and identified the workflows that absolutely could not regress. That list became the spec for the replatform — and the test plan.
From there we built the new system in a parallel slice. Patient records and appointments mirrored from the old database in near-real-time. Schedulers could toggle between the two systems on a per-clinic basis, which meant we could roll forward — and back — clinic by clinic, never the whole organization at once.
The hard parts
The hardest decisions weren't technical. They were about what to not port over. The legacy system had ten years of accumulated features, half of them used by one clinic each. We worked with operations to retire the ones that no longer earned their keep, and to design clean replacements for the ones that did.
A second hard part: keeping the scheduling team confident through the transition. We did weekly demos with them, kept a public migration tracker, and made sure every change they reported was acknowledged within a day.
Outcome
The replatform took just over six months. By the end:
- All twelve clinics were on the new system.
- Deploy time went from sixty minutes to under three.
- The on-call rotation went from a weekly thing to a monthly thing.
- The internal engineering team — three people — was confidently shipping changes on their own within a week of handoff.
The system that handled the migration is the same system the team uses today, with the same code patterns. We left, and it kept running.