AI translation QA scores each translated segment for likely quality without needing a reference translation, and runs targeted checks for terminology, numbers, tags, omissions and meaning shifts, so reviewers spend their time on the segments most likely to be wrong.
All Translation & Localisation →
Classic QA checks in Xbench or the CAT tool are good at form: inconsistent repetitions, mismatched numbers, broken tags, double spaces. They cannot see meaning. A dropped “not”, a sentence that is fluent and says something else, a register wrong for the audience: those pass every rule and are precisely the errors that matter in a contract or a safety instruction.
Machine translation quality estimation fills that gap. Models trained on human quality judgements, and language models prompted with an error typology such as MQM, can flag segments likely to contain accuracy errors, and say what kind. Neither is reliable enough to sign off a translation, and both have blind spots by language pair. Used to order review rather than replace it, they move attention to where it is needed.
We calibrate on your own reviewed jobs: segments your linguists already corrected, with the corrections as ground truth. For each language pair we measure how many of the segments that needed real correction land in the high-risk group, and how many clean segments are flagged unnecessarily. Those two numbers set the thresholds, and they are rechecked by sampling the low-risk group on live jobs, because a quality estimator that drifts quietly is worse than none.
Quality estimation predicts how good a machine translation is without a human reference translation to compare it with. It scores segments, sometimes highlighting the words likely to be wrong, so reviewers can prioritise. Research systems such as the COMET family and prompted language models both do this, with accuracy that varies by language pair and domain.
Not for content where errors have consequences. AI QA is good at directing attention, finding the segments most likely to be wrong and catching mechanical errors consistently. It misses things, and it has no accountability for the result. The reliable pattern is AI to prioritise and pre-check, and a qualified reviewer to decide.
It complements them. Rule-based checks remain the right tool for form errors and we keep them, or run equivalent checks in code. Quality estimation adds a view of meaning that rules cannot provide. Findings are delivered as comments or flags in the CAT tool, so reviewers work where they already are.
What we built, what stage it is at, and a link to the live thing wherever there is one.
yourOS is an operating system with an AI at its heart. You talk to it like a person, out loud or typed. It makes your calls, plans your days and works on your projects with a crew of agents, on your own private server. Our own product, built from the first commit; the waitlist is open.
Website, backend and app assets for Sonora, a free AI sound-therapy app on iOS and Android published by Cognitive AI.
Brand, website, prototype, apps on five platforms, the AI backend and the algorithm-intelligence pipeline for Podswap, a social growth engine that runs without a human in the loop.
In closed beta: Roxy, a private AI assistant with permanent memory running on infrastructure the user owns. Brand, site, product, backend and conversational design.
Live at coherence.com: brand identity generated with AI agents we directed, the motion system and website for the group behind Sonora, AboutAI, WakeUp and MudShadows. Agent work remains in build.
Live at domainbooth.com for DomainBooth, a premium domain brokerage: a new animated mark, every page redesigned and rebuilt with designed mobile layouts, and a custom CMS for its sales record, featured names, press and enquiries.
We calibrate on jobs your reviewers already corrected, pair by pair.
30-minute technical call · no deck