localization-urdu - SKILL.md Agent Skill

name: localization-urdu description: Provide per-chapter Urdu translation toggle and manage translated content delivery. Use when translating chapters, caching translations, or implementing language toggles.

Localization Urdu Skill

Instructions

Translation pipeline
- Extract sections by heading from source Markdown/MDX
- Translate via OpenAI/ChatKit translation model
- Keep glossary for robotics terms (JSON file)
- Cache translations with checksum per section
- Rerun only when source checksum changes
Storage
- Option A: Markdown copies (.ur.mdx) alongside English
- Option B: Neon table translations(section_id, lang, content)
- Tag Qdrant payload with lang for bilingual RAG
Frontend
- Add language toggle in Docusaurus layout
- Load Urdu content when available, fallback to English
- Ensure chat widget passes desired language to backend
Quality
- Spot-check key technical terms
- Allow glossary overrides via JSON config
- Keep technical terms consistent across chapters

Examples

# Translation function
async def translate_section(content: str, glossary: dict) -> str:
    prompt = f"""Translate to Urdu. Keep technical terms from glossary.
    Glossary: {glossary}
    
    Content: {content}
    """
    # Call OpenAI
    return translated_content

// glossary.json
{
  "Physical AI": "فزیکل اے آئی",
  "Humanoid Robot": "ہیومنائیڈ روبوٹ",
  "ROS 2": "ROS 2",
  "NVIDIA Isaac": "NVIDIA Isaac"
}

Definition of Done

Translation pipeline runs and caches results; reruns are incremental
Urdu toggle renders translated sections for at least one sample chapter
Backend can return Urdu-context answers when requested