Diagram read best practices

[Murat1990]

Hi team,

I'm working on improving our threat modeling application's diagram processing capabilities. Currently facing challenges with:

1. OCR limitations in accurately extracting information from architecture diagrams
2. Inconsistent results when processing PDF/JPG/PNG formats
3. Loss of structural information when converting diagrams to text

I've heard DSL (Domain Specific Language) might be a better approach. Looking for recommendations on:

• Alternative approaches to diagram processing beyond OCR
• Best practices for maintaining diagram structure during processing
• Tools/libraries for parsing architecture diagrams programmatically
• Experience with DSL-based solutions (pros/cons)

Current stack:

• Python/AWS Bedrock/Claude
• Image processing: Pillow, pytesseract
• Input formats: PDF, JPG, PNG

Would appreciate any insights from those who've tackled similar challenges.

[Julian-Dumitrascu]

1. We can have this conversation, as we could help you make such a choice.
   1.1 What has happened in the meanwhile?
2. What is topical for you nowadays?