i am working on a open ai assistant which function is to validate a series of contracts based on how many checkboxes are checked.
However, it seems difficult to the assistant to do this task.

I have been trying two options:

• Uploading it as a scanned pdf (unselectable words) and making the assistant recognize the text by OCR. This sometimes work, but with very low percentage of success.
• Uploading the pdf as it is. With this, the assitant is capable of getting all the text and data, but it cannot detect the checkboxes.
  Do you happen to know any possibility, or combination with other technologies that can help me solve this problem?
  Thank you in advance.