Two minutes off the slowest workflow
Farakav, an enterprise process-mining platform · Golrang System · 2025 to present
- React
- TypeScript
- Vite
- Ant Design
- TanStack Query
Outcome
- Up to two minutes removed from the slowest critical workflows
- Hundreds of per-field queries removed from a single request path
The problem
Several workflows on the platform were slow enough that people had learned to start them and go and do something else. Slow in a way that everyone had stopped reporting, because it had always been like that.
Finding it needed no instrumentation and no tooling. One endpoint was visibly slower than everything around it, and the network panel was enough to say so with confidence. The question worth asking was not which request was slow. It was what that request was doing.
What I did
It was running a query for every field it returned, and a response could carry hundreds of fields. All of that work produced data that was only ever displayed once a user clicked an individual item — which, for most items, nobody did. So I moved it to the click. The list request returns the list; the detail query runs once, for the one item someone actually opened. Hundreds of queries per request became none.
What I would do differently
I accepted the shape of that endpoint for far longer than I should have. It had been slow since before I arrived, everyone had stopped mentioning it, and I treated that as a fact about the system rather than a question I was allowed to ask. Once I asked it, the fix was small. What it cost was all the time nobody spent asking.