AI translation has reached commercial quality for many language pairs. Stack choices: 1) DeepL API: highest quality for European languages, domain-specific glossaries, excellent API. ~$0.03/1K chars. 2) Google Cloud Translation (V3): broadest language support (100+ languages), good for high volume. AutoML translation for custom domains. 3) LLM-based (GPT-4o/Claude): best for context-aware translation, marketing copy, idiomatic expressions. 10-50x more expensive than specialized models but handles nuance better. 4) Custom domain models: fine-tune MarianMT or mBART on domain-specific parallel corpora for technical/legal translation. Quality evaluation: BLEU score (automated, compares with reference translation), COMET (neural, better correlation with human judgment), BERTScore, human evaluation for high-stakes content. Localization beyond translation: adapt formatting (dates, currencies, numbers), cultural references, UI layout for RTL languages, images and icons with cultural sensitivity. Workflow: auto-translate -> quality estimation (COMET quality estimation model identifies low-confidence segments) -> human review for low-quality segments -> post-edit -> back-translation verification. For software: use i18n frameworks (react-i18next, vue-i18n), extract strings automatically, maintain translation memory to reuse previous translations.