How to build a Form Processing Agent

Workflow Overview

1. Upload Handwritten Form

• Node: Upload Handwritten Form (doc-0)

• What it does:
  The user uploads a scanned or photographed handwritten insurance claim form (e.g., First Report of Injury, Property Loss Notice, General Liability Incident Report).

• How it works:
  This node uses OCR (Optical Character Recognition) to extract text from the uploaded image or PDF, making the handwritten content machine-readable for downstream processing.

2. Data Extraction

• Node: Data Extraction (llm-0)

• What it does:
  An AI model (LLM) reads the OCR text and extracts all key fields from the form, such as:

  • Claimant name

  • Date of loss

  • Policy number

  • Incident description

  • Contact info

  • Signatures

• Special features:

  • Outputs the extracted data as structured JSON.

  • If a field is missing or illegible, it sets the value to null and adds a note in a flags array (e.g., "missing date", "blurry signature").

3. Validation and Formatting

• Node: Validation and Formatting (llm-1)

• What it does:

  • Checks the extracted fields for completeness and clarity.

  • Flags any missing or illegible data.

  • Reformats the information into a clean, readable Markdown document for insurance adjusters.

  • Adds a "flags" section at the end if any issues are found.

4. Claimant Name Extraction

• Node: Claimant Name (llm-2)

• What it does:

  • Specifically extracts just the claimant’s full name from the structured data produced by the Data Extraction node.

  • This is used to personalize the file name of the generated Google Doc.

5. Create Google Doc

• Node: Create Google Doc (action-1)

• What it does:

  • Creates a new Google Doc in your Google Drive.

  • The document’s content is the validated and formatted Markdown summary from the previous step.

  • The file is named after the claimant (e.g., "John Doe Form").

• Connection:

  • Uses your Google Drive connection to create and store the file.

6. Output

• Node: Output (out-0)

• What it does:

  • Presents the final result to the user, typically showing a link to the newly created Google Doc and a summary of the processed claim.

Summary Table

How the Flow Works in Practice

1. User uploads a handwritten claim form.

2. OCR extracts the text.

3. AI extracts all relevant fields and structures them as JSON.

4. A second AI step validates the data, flags issues, and formats it for adjusters.

5. The claimant’s name is extracted for use as the document title.

6. A Google Doc is created with the formatted summary, named after the claimant.

7. The user receives a link to the Google Doc and a summary of the processed claim.