Hi,
I’m engaged in a Java project aimed at converting scanned PDF documents of physical agreements into Java objects for further analysis and processing.
Despite providing clear and specific instructions to GPT for identifying JSON field locations within these documents, I’m facing significant challenges with conversion accuracy. About half of the output is accurate and aligns with expectations, but the remaining portion includes substantial errors, misreads, and occasionally triggers internal server errors from OpenAI. Notably, all PDF pages have been converted into high-quality, uncompressed images of printed (not handwritten) text.
Given the high clarity of these images, the output is error-prone and inconsistent. If anyone has suggestions on improving text recognition accuracy or any strategies that could help restore confidence in this project, I would be immensely thankful for your advice.
Also, if anyone has encountered similar issues or has experience with related projects, hearing about your challenges and solutions would be greatly beneficial.
Thank you for your time and thoughts!