How to build a Form Processing Agent

This agent enables users to upload scanned or photographed handwritten insurance claim forms (such as First Report of Injury, Property Loss Notice, or General Liability Incident Reports). The uploaded file is processed using OCR to extract the raw text, which is then analyzed by a large language model (LLM) to identify and pull out all relevant structured data fields (e.g., claimant name, date of loss, policy number, incident description, contact info, and signatures). The extracted data is validated and reformatted into a clean, readable Markdown summary for adjusters, with any missing or illegible fields flagged for review. The claimant’s name is also extracted separately for use as the document title. Finally, the agent automatically creates a new Google Doc in the connected Google Workspace account, containing the formatted summary and a link for easy access and record-keeping.