Hello,
I am using Gemini API models to extract information from PRD (Product Requirement Document) files. Our goal is to extract all relevant details in text form, including text information from images, flowcharts, descriptions, tables, diagrams, annotations, and any other structured or unstructured elements within the document.
However, we are encountering the following issues:
- Inconsistent Data Extraction:
Some flowcharts, tables, or diagrams are not fully processed or extracted. The text information from certain sections (like images, annotations, or descriptions) is incomplete or missing entirely.
- Skipped Content in Longer Documents:
For longer PRD documents, the models often fail to read the complete content and skip significant portions of the document. Has anyone experienced similar challenges with Gemini API models? Are there any recommended configurations, preprocessing techniques, or strategies to ensure the models process the entire document and improve the accuracy of the extracted data?
Any guidance, best practices, or workarounds would be greatly appreciated.
Thank you!