For years, the post-editing step in translation workflows was the part nobody had fully solved.
Machine translation kept getting better. Linguists kept getting faster. But the review step — the moment a human had to make sense of AI output before anything could go to a client — stayed slow, opaque, and difficult to standardize. You never quite knew what you were dealing with until you were already in the middle of it.
We built re[words] because we got tired of working around a gap that shouldn’t exist.
The problem wasn’t the MT. It was the visibility.
When a translated file comes out of an MT engine, it looks finished. That’s the problem. Fluent output and correct output are not the same thing — and a surface reading won’t tell you which one you’re looking at.
A segment can be grammatically sound and still miss a glossary term. It can read naturally in the target language and still get the locale convention wrong. It can feel accurate and still drop a formatting tag that will break the final file.
Post-editors know this. They’ve always known it. The question is how long it takes to catch these issues — and whether there’s any systematic way to flag them before a human has to find them one by one.
That’s the gap we built an AI post-editing tool to close.
What re[words] actually does
Re[words] takes your XLIFF files — the format your CAT tool already works with — and runs them through a structured quality analysis designed by professional linguists. It returns two things: a corrected file, and a processing report that documents the reasoning and changes applied across six quality dimensions.
The linguist doesn’t arrive at a raw MT file and start from scratch. They arrive at output that has already been assessed, with a clear map of what was flagged and why. The review step still happens — that’s by design — but it starts from a much better position.
Sure, you could prompt a model directly. But there’s only so much you can do without the right setup behind it. Re[words] is configured around your content from the start: our AI language specialists work with you to review your terminology glossary and build custom prompts tailored to your register, brand voice, and domain requirements. The configuration lives in your account, is yours exclusively, and is a standard part of every subscription — not an optional add-on.
The decisions that shaped it
A few things were non-negotiable from the start.
Data privacy had to be a first principle, not an afterthought. Re[words] connects via API — your content doesn’t pass through our servers. If you bring your own API key (BYOK), the data flow is entirely between your machine and the model provider. The API connections are subject to enterprise data policies that don’t use your content for model training by default.
The quality dimensions had to be independently toggleable. Not every file needs every check. A client with a tight glossary and a regulated terminology requirement needs a different configuration from a client publishing high-volume support content. The tool needed to reflect that.
And it had to work with existing workflows, not replace them. Re[words] is XLIFF-native. It supports SDL Trados, Phrase, memoQ, and Okapi Framework. You don’t change how you work — you add a step that makes the step after it faster.
What it isn’t
Re[words] is an AI post-editing tool. It is not a translation service, and it does not replace human judgement.
Human review remains a mandatory step in any responsible workflow. The tool is designed to make that review faster, more structured, and more traceable — not to skip it. Responsibility for the final output sits with the professional or agency delivering it. That’s not a disclaimer. It’s how we think professional workflows should work.
It’s also not a generic AI wrapper. The quality logic was built specifically for translation, by people who have spent years doing it.
Why we’re talking about it now
Re[words] is live.
If the problem we’ve described sounds familiar — if your post-editing step is slower than it should be, or your reviewers are catching the same categories of errors file after file with no systematic way to address them — book a demo and we’ll walk you through it.
