AWS Textract PDF Form Parser Demo

This application demonstrates how to use AWS Textract's AnalyzeDocument API (FORMS feature) to parse PDF forms, extract structured Key-Value pairs (including checkboxes), and display them in an Angular frontend.

Technology Stack

• Frontend: Angular 21+ (Standalone Components), TypeScript, CSS (Flexbox/Grid).
• Backend: Node.js, Express.js.
• AWS Services:
  • Amazon Textract: Specifically the StartDocumentAnalysis API with FeatureTypes: ["FORMS"] for structured data extraction.
  • Amazon S3: For temporary storage of documents required by the asynchronous Textract API.
• Libraries:
  • @aws-sdk/client-textract: For interacting with Textract.
  • @aws-sdk/client-s3: For uploading files to S3.
  • multer: For handling file uploads in Express.

Architecture & Workflow

The application follows a standard asynchronous processing flow to handle potentially multi-page PDF documents:

1. File Upload:

   • The user selects a PDF in the Angular frontend.
   • The file is sent to the Node.js backend via the /analyze endpoint using FormData.

2. S3 Staging:

   • The backend uploads the received file to a designated Amazon S3 Bucket.
   • This step is necessary because the asynchronous StartDocumentAnalysis API requires the input document to be in an S3 bucket (it cannot accept raw bytes for large files).

3. Textract Analysis Job:

   • The backend initiates a Textract job using StartDocumentAnalysisCommand.
   • Key Configuration: FeatureTypes: ["FORMS"] is enabled. This tells Textract to specifically look for key-value pairs (e.g., "Name: John Doe") and selection elements (checkboxes).

4. Polling Validation:

   • The backend polls the GetDocumentAnalysisCommand API every 1 second to check the Job Status.
   • Once the status changes to SUCCEEDED, the backend retrieves all result pages (handling pagination/NextToken).

5. Data Parsing & Structured Output:

   • The raw Textract JSON response consists of thousands of "Blocks" (PAGE, LINE, WORD, KEY_VALUE_SET, SELECTION_ELEMENT).
   • Data Parsing & Schema Validation:
     • Fuzzy Matching: extracted keys are compared against a canonical generated_schema.json using Levenshtein distance. This corrects OCR typos (e.g. Sentenc Date -> Sentence Date) and ensures consistent JSON keys.
     • Boolean Extraction: The parser specifically detects fields that contain checkboxes (Selection Elements). Even if a field contains mixed text (e.g., [ ] (Less than...)), the logic extracts the boolean state (true/false) while ignoring the noise.
     • Pruning: Extraneous page elements (Footers, Page numbers) defined in the schema are automatically filtered out.
   • Schema Enforcement: generated_schema.json acts as a strict contract, defining not just the keys but the types (e.g., boolean for checkboxes). This ensures the frontend receives predictable, type-safe data.

6. Response:

   • The backend returns a unified JSON object containing both the raw Blocks (for the visual overlay) and the cleaned StructuredData (for the form table).
   • The Angular UI renders the text overlay on the PDF and populates the data table.

API Endpoints

POST /analyze

• Purpose: Extracts structured form data with schema validation.
• Input: multipart/form-data with file (PDF).
• Output: JSON object { "StructuredData": { ... } }.
• Details: Returns simplified, normalized key-value pairs matching generated_schema.json.

POST /analyze-blocks

• Purpose: Returns raw Textract blocks for visual overlays.
• Input: multipart/form-data with file (PDF).
• Output: JSON object { "Blocks": [ ... ] }.
• Details: Returns the full hierarchy of PAGE, LINE, WORD keywords and bounding boxes.

Deployment & Configuration

Prerequisites

1. Node.js: v24+ installed.
2. AWS Account: Active account with permissions to manage Textract and S3.

AWS Configuration

1. IAM Permissions: Ensure your user has AmazonTextractFullAccess and AmazonS3FullAccess.
2. S3 Bucket: Create a bucket (e.g., textract-demo-bucket) in your region (e.g., us-east-1).

Environment Variables

Create a .env file in the root folder (textract_sample/.env):

AWS_ACCESS_KEY_ID=your_access_key_id
AWS_SECRET_ACCESS_KEY=your_secret_access_key
AWS_REGION=us-east-1
BUCKET_NAME=your_existing_s3_bucket_name

Running the Application

Terminal 1: Backend

cd textract_sample
npm install
node server.js

• Server starts on port 3001.

Terminal 2: Frontend

cd textract_sample/textract-demo
npm install
ng serve

• Access the app at http://localhost:4200.