Replacing two SaaS tools without anyone missing them
The internal operations portal of a digital-services consultancy, for time, tasks and website audits.
- Role
- Lead engineer; took it over before launch
- Time
- 8 months
- Stack
- Next.js · Postgres · Prisma
- Team
- A previous developer laid the first foundation
Problem
A growing team split its day between a third-party time tracker and a third-party task manager, and its website audits struggled with large sites.
Constraints
- The whole team moves at once, and its history has to come along.
- Nobody can lose an hour of tracked time in the move.
- Audits have to fit on the machines that already run them.
Decisions
- Moved the time history with an importer that is a dry run by default, insert-only and idempotent, verified per person and month, and refuses a partial pull, because a half-imported history is worse than none.
- Attributed and froze hours when they are written instead of recomputing them later, with a database rule that allows one running timer per person.
- Took project-wide row locks in a fixed order when tasks move, so concurrent reparenting cannot create cycles or deadlocks.
- Split each audit into child tasks, one Lighthouse run per machine and a second wave for leftovers, so large sites finish.
- Moved reads from Server Actions to GET handlers after finding that Server Actions run one at a time, and made timers optimistic so the screen never waits.
Outcome
The whole team moved off both tools, and audits now cover twice as many pages per site.
- 200 → 400 pages per audit
- About 3,600 commits in 8 months
Credits
The first stretch of the project (the original scaffold, a framework port and the first audit tool) was a previous developer's work.