Humanitarian document workflow from content extraction and AI translation to review and email delivery

The challenge

During active storm response, leadership sent unstructured Vietnamese PDFs, scans, and images for English-language updates. Monitoring and reporting could recur three to nine times a day.

My role

I developed the assisted document-preparation workflow and independently conceived and built a separate automation pilot. A Vietnamese colleague checked translations before leadership distributed the updates. I presented both approaches to the United Nations Development Coordination Office headquarters in New York.

Operational support and a separate pilot

  1. Prepare live updates

    Extract text from unstructured Vietnamese documents with OCR, use generative AI to support translation, and format English-language updates.

  2. Keep language review human

    Send the draft to a Vietnamese colleague for review before leadership shares it with the UN disaster risk reduction team in Viet Nam and UN Viet Nam.

  3. Prototype automation

    Build a React file-upload interface connected to an n8n Webhook, OCR node, and local AI model to explore extraction, translation, formatting, and routing.

Tools & Methods

Tools and technology

  • OCR
  • Generative AI
  • React
  • n8n
  • Local AI

Methods

  • Document processing
  • AI-assisted translation
  • Human review
  • Workflow prototyping

Pilot boundary

The automation remained a pilot. Its translation accuracy had not been validated sufficiently for operational use.

Continue reading

A virtual patient for practical clinical learning

View next project