Handling Large PDF Files in Logic Apps + GPT Workflow

I have built an Azure Logic Apps workflow that:

1. Monitors a SharePoint location for new PDF files.
2. Extracts text from the PDF.
3. Sends the extracted content to a GPT agent/persona for information extraction and summarization.
4. Receives the structured response in JSON format.
5. Parses the JSON and writes the output to a SharePoint-hosted Excel table.

The workflow works well for most documents. However, it fails for certain PDFs that contain a large number of embedded images or have a large overall file size, resulting in token/context limit errors when sending the document content to the GPT persona model.

To address this, I also experimented with parsing the document and splitting it into smaller token-based chunks before sending the content to the GPT persona. Unfortunately, the process still fails during the document parsing stage itself for some large or image-heavy PDFs, before chunking can even occur.

I'm interested in learning how others have addressed similar challenges. Specifically:

• Best practices for processing large contract documents that exceed model token limits.
• Approaches for handling image-heavy PDFs.
• Chunking or section-based processing strategies that preserve document context.
• Architectures that work well with Logic Apps and Azure OpenAI for large-document extraction workflows.

Any recommendations, lessons learned, or reference architectures would be greatly appreciated. Thanks!