drewgillson / googlepalm-minute-book-extraction

Extract information from corporate minute books: details about the corporate entity, directors and officers, and clauses from the shareholder's agreement like quorum rules, restrictions or provisions, and share classes.

Summary

This Google Cloud solution uses Document AI Custom Document Classifier, OCR Processor, Form Parser, and the Google PaLM API to extract information from corporate minute books: details about the corporate entity, directors and officers, and clauses from the shareholder's agreement like quorum rules, restrictions or provisions, and share classes.

Solution Overview

• Splits each page from a multi-page PDF into individual pages and saves PDFs to Cloud Storage
• Classifies pages using a Custom Document Classifier trained to distinguish types dense-ocr, form-parser, certificate, or other
• Parallelizes text extraction with Cloud Function instances that invoke Document AI Processors based on the page type
• Augments OCR text with output returned from the Document AI Form Parser processor for form-parser pages
• Steps through each page of OCR text to collect relevant entities into the extraction schema using heuristics and LLM prompts
• Writes structured JSON output with entitities of interest to Cloud Storage

Requirements

• Google Cloud project with a Cloud Storage bucket, Document AI OCR Processor, Form Parser, and Custom Document Classifier
• Terraform v1.4.5 to deploy Cloud Functions, Pub/Sub queues
  • Update terraform/modules/base/outputs.tf with your own instance IDs