Retiring a homegrown MT tool, keeping the best bits (2024)

Background

During lockdown I built a small app that automated translation for awkward file types and ran on my own VM. It solved a real need at the time, but over the years TMS connectors and APIs caught up. The right move was to retire the public tool while keeping what still helped on bespoke projects.

What I built

A migration path that moved teams to stable TMS connectors for common formats and kept a slim ‘sidecar’ for the edge cases—files with odd structures, strict placeholder rules, or unusual round‑trips.

Clear guidance that AI/MT is a helper, not the whole answer: use it to remove routine work, keep decisions with people, and run simple checks that catch mechanical errors early.

How we ran it together

PMs used the connector flow for normal content; the sidecar handled the few tricky formats and produced clean packages either way. Translators worked in the same tools; engineering saw the same shape back.

Outcome

Less to maintain, fewer surprises, and a straighter path for everyday projects, with a safety valve for the oddball cases where a tailored pipeline still pays for itself.

If I were starting this today

I’d make the same choice and add small, explainable AI checks around numbers and placeholders—useful, visible, and easy to switch off.