Localization

Extract user-facing strings from a codebase, translate them into target languages, and verify the translations render correctly in context.

Workflow

runPipeline(
  extractStrings
    .then(iterate().map(
      all(...targetLanguages.map(lang => translateTo(lang)))
        .then(iterate().map(verifyInContext).collect()),
    ).collect())
    .then(flattenResults)
    .then(writeLocaleFiles),
);

Stages

1. Extract strings — scan the codebase for user-facing text. Output: array of { key, text, context }.
2. For each string (concurrently):
   • Translate — all runs translations into every target language concurrently.
   • Verify — check each translation in context (e.g., does it fit the UI, is it culturally appropriate).
3. Write locale files — aggregate all translations and write them to the appropriate locale files.

Key points

• Translation into different languages is embarrassingly parallel — all runs them concurrently.
• The verification agent sees the translation, the original text, and the UI context — but not other translations.
• The extraction step is deterministic (AST parsing or regex), not an LLM.
• .iterate().map() processes all strings concurrently, so the workflow scales with the number of strings.