I’m looking to develop an automation process to extract specific information from 15 to 30 pages within PDF documents. The challenge is that the required values are not consistently located on the same page or in the same place. Using a PDF parser tool like pdf.co to extract this information might become expensive as I’d need to parse the entire file.
I’ve considered uploading the file to OpenAI in order to train an agent to extract this information; however, it seems that OpenAI doesn’t have direct access to the PDF and provides outputs based on the examples I provide. I also attempted to convert the full PDF to text, but it becomes too large to create a prompt for OpenAI or other similar platforms. When I divided it into chunks, I encountered difficulties in using all the chunks to prompt the agent.
Does anyone have any suggestions on how to create a solution for this, or how to improve the interaction with OpenAI’s assistance?
I did this but the assistant doesn’t reply based on the content I submit but with the examples I provided in the instructions. Do you know what can be wrong?
I dont know. The setup is exactly like yours.
The pdf is in the vector store, and I do have enough credits in my account. I selected the right vector store and the assistant was created with gpt4o. But then the output is not right
If you give the same input and pdf in playground do you get the output you want?
Essential if the playground works, then all you are doing via make is automated it, and the hardest step was for it to find the pdf. Which you say is there.
To accurately parse specific pages from your PDF file using the PDF.co Document Parser, please use the pages parameter. By specifying the pages you want to extract, the Document Parser will focus only on those pages instead of processing the entire file. For more details on how to use the pages parameter, please visit our documentation at the following link: API Docs
If you have any questions or need further assistance, please let us know.
