Reading articles in a unfamiliar language is a pain. I’m documenting the working that worked for me to read a multi-page Dutch article. Digitizing physical articles for translation is cumbersome. This workflow leverages Mistral AI’s Document AI features to handle OCR and translation efficiently.

The Digitization Workflow

1. Capture and Upload

Start by taking clear photos of the article pages.

• Upload: Use the Mistral AI Studio Document AI Playground.
• Processing: Upload the images to extract text. Note that if you have more than one page in a single photo frame, you may need to spot-check the output.

2. Annotation and Cleanup

• Page Numbers: Manually insert page numbers at the top of the generated markdown files to maintain order during the merge process.
• Editing: Correct any OCR errors directly in the playground before downloading.

3. Merging Markdown Files

Once the markdown files are downloaded for each page, use a script to combine them. This ensures the document stays continuous and coherent.

• Script: Run a Python script to combine files in alphabetical/numerical order.
• Reference: Combination Script Gist

4. Translation

Take the combined_markdown.md file back to the Mistral AI Studio.

• Goal: Use the large context window and high reasoning capabilities of Mistral models to translate the entire Dutch article into your target language.
• Benefit: Translating the combined file (rather than individual pages) helps the AI maintain context and terminology consistency across the whole article.

This method transforms a stack of physical papers into a searchable, translated digital asset in minutes.