Technology you can trust with documents
Translation memory, glossaries and a cloud language model — in one loop, under the translator's control.
The processing loop
The document is split into segments. Exact matches and repetitions are covered from translation memory; the remainder goes to the cloud language model as separate fragments — no file, no credentials. Files, memory and glossaries stay on servers in Kazakhstan.
Context for every fragment
Each fragment is paired with the client's glossary, memory matches, neighbouring sentences and domain translation rules — so the model translates a term the way your glossary does, not the way a dictionary does.
Checks and confirmation
Numbers, units, tags and terms are checked before confirmation. A confirmed segment feeds the memory and is reused in the next documents — repeated work is not paid for twice.
Translation rules per client
Translator edits become proposed translation-rule changes confirmed by a person. Rules are versioned; rollback takes one click.
Kazakh quality and industry corpora
Industry translation rules, terminology control and memory built on medical and legal texts. Kazakh is our core specialty.
How the translation pipeline works
From uploading a file to exporting the finished document.
Document → segments
The file is split into segments. From here the system works with fragments, never the whole dossier.
Draft: memory first, then AI
Exact matches from translation memory are inserted first. The remainder goes to the cloud language model as separate fragments — no file, no credentials.
Context: glossary and rules
Each fragment is paired with the client's glossary, memory matches, neighbouring sentences and domain translation rules.
Human review
A translator confirms segments with quality checks: numbers, units, tags. Confirmed segments feed the memory and are reused in the next documents.
See VOXA on your own documents
A 20-minute demo — and you'll see the editor, the memory and the Kazakh quality on your own examples.