I have multiple PDF invoices in various formats, including handwritten invoices, And I need to extract key information using AI/LLM tools. Currently, I am using the Generate Text Completion activity in UiPath to connect to Amazon Bedrock, leveraging the latest model (Anthropic Claude Sonnet 4.5) for information extraction.
However, the model does not accept PDF files directly. As a workaround, I first convert the PDF into a text file and then pass the extracted text to the LLM via a prompt to retrieve the required invoice details.
I would like to know if there is an alternative approach that allows me to process PDF files directly, without having to convert them to text file first—either through Bedrock-supported capabilities or a different UiPath integration.
Currently, UiPath’s Generate Text Completion activity can’t send PDF files directly to Amazon Bedrock, so what you’re seeing is expected behavior.
The usual workaround is to use Amazon Textract or UiPath Document Understanding first. Textract can read scanned and handwritten invoices and gives much better structured output than a simple PDF-to-text conversion. That output can then be passed to Bedrock for interpretation.
Bedrock does support multimodal inputs in some APIs, but UiPath hasn’t exposed that capability yet. To use it, you’d need a custom API or Lambda integration.
I think Bedrock LLM models cannot take PDFs directly, so text extraction first is required. If you want to process PDFs without converting them to text yourself, use UiPath Document Understanding with OCR or AI Center models because they accept PDFs natively. Bedrock currently has no direct PDF input support through UiPath.
If helpful, mark as solution. Happy automation with UiPath
You should try leveraging UiPath Agents for this purpose. They support file as a direct input and you can choose from a pool of different LLM models. You will just have to add an inbuilt tool named ‘Analyze Files’ to leverage model for data extraction from the input file.
