While evaluating whether llama_parse would work for our use case, I noticed that llama_index appeared to ignore a large portion of the text in the test document I used.
When I opened said document using LibreOffice Draw, I noticed that the text was in the form of Bézier curves.
While evaluating whether llama_parse would work for our use case, I noticed that llama_index appeared to ignore a large portion of the text in the test document I used.
When I opened said document using LibreOffice Draw, I noticed that the text was in the form of Bézier curves.
Here is a minimal reproducible example: exhibit_a.pdf exhibit_b.pdf
exhibit_a.pdf
contains two regular text frames, andexhibit_b.pdf
contains one text frame converted to Bézier curves, and one regular text frame.Here is the output for![image](https://github.com/user-attachments/assets/4a545e34-7fc5-4c7c-8005-849f45438fca)
exhibit_a.pdf
:Here is the output for![image](https://github.com/user-attachments/assets/3909b813-05ff-457a-8417-d30eb662c20a)
exhibit_b.pdf
: