SYSTEM Cited by 1 source
Atlassian Localization Pipeline¶
Atlassian's Localization Pipeline is the production workflow that keeps translation throughput aligned with AI-era feature delivery. It combines pretranslation by an internal AI system or external AI vendor hub, mandatory professional-translator approval in Smartling, and developer-facing source-message validation. Its defining design choice is to improve both ends of the flow: make source text translatable before intake and give translators context-rich drafts rather than raw strings. (Source: sources/2026-08-06-atlassian-scaling-localization-at-atlassian-keeping-translation-at-the-pace-of-ai-era-development)
Architecture¶
- Source validation: editor and code-review tooling identifies untranslated text, English-only string composition, missing plural rules, and insufficient context.
- Context assembly: each request receives developer notes plus similar prior translations, glossary terms, and locale style rules.
- AI pretranslation: an internal Atlassian system or external vendor hub writes the first draft.
- Human release gate: a professional translator reviews and approves every string in Smartling before release.
- Learning loop: reviewer edits are retained as feedback for future drafts; Translation Edit Rate tracks the distance between draft and approved version.
Operating model¶
The system supports products in more than 20 languages. After AI-assisted development raised translation input volume 272% year over year in FY25 H2, the pipeline used pretranslation to make reviewer time concentrate on meaning, tone, and local fit. It does not treat low edit distance as sufficient evidence to bypass translators. (Source: sources/2026-08-06-atlassian-scaling-localization-at-atlassian-keeping-translation-at-the-pace-of-ai-era-development)
Quality and remediation¶
Atlassian uses the same source-quality rules preventively and retrospectively. Checks guide a developer in the editor and at code review. For accumulated frontend defects, AI performs repetitive remediations in bulk and creates pull requests for developer review. The post reports more than 20,000 existing issues cleared, but does not publish precision or merge-rate data. (Source: sources/2026-08-06-atlassian-scaling-localization-at-atlassian-keeping-translation-at-the-pace-of-ai-era-development)
Seen in¶
- sources/2026-08-06-atlassian-scaling-localization-at-atlassian-keeping-translation-at-the-pace-of-ai-era-development — canonical public description; model/vendor internals and validation-rule implementation are not disclosed.