AWS Bedrock pipeline extracts data from scanned docs on-demand or in batch
Unlock business intelligence from millions of PDFs with flexible AI pipelines on Amazon Bedrock.
Amazon Bedrock can be used to build an intelligent document processing pipeline with on-demand and batch inference options. The on-demand pipeline uses an SQS FIFO queue and Lambda to extract data from scanned PDFs within seconds by converting pages to images and calling multimodal large language models. According to the article, Claude 4 Sonnet allows a maximum of 20 images per invocation, so larger documents are split. The batch pipeline processes multiple document requests asynchronously for cost efficiency. Users can dynamically specify the LLM model ID and prompt per document, enabling extraction from varied document types using the same pipeline.
- On-demand pipeline uses SQS FIFO queue + Lambda to return extracted data in seconds for time-sensitive requests.
- Batch pipeline processes multiple documents asynchronously via Amazon Bedrock, reducing cost for large backlogs (e.g., hundreds of millions of land lease PDFs).
- Each document can specify its own LLM model ID, prompt ID, and version, enabling flexible extraction across diverse document formats.
Why It Matters
Unlocks structured data from massive volumes of scanned documents at scale, balancing speed and cost.