LLM Document Parser

This Blueprint provides a locally runnable pipeline for parsing structured data from scanned or digital documents using open-source OCR and LLMs. It takes in one or more documents in image and/or PDF formats as input and returns a single structured object with fields parsed from the documents. By defining a prompt and data model, the Blueprint will know how what fields to parse and what they should look like.

The example use case, parsing transaction data from bank statements, demonstrates how you can pass in multiple documents with differing formats and extract shared fields (transaction amount, description, and date). All of the bank statments from every document are compiled into one object. This Blueprint can be customized to work with any type of document to fit your needs.