Document Redaction

Original Document

Redacted Document

Example of document redaction applied to a medical form

from vlmrun.client import VLMRun
from vlmrun.client.types import FileResponse
from pathlib import Path

# Initialize the client
client = VLMRun(api_key="<your-api-key>")

# Upload the file
response: FileResponse = client.files.upload(
    file=Path("<path/to/sensitive_document.pdf>")
)
print(f"Uploaded file:\n {response.model_dump()}")

Uploaded file:
{
  'id': '1e76cfd9-ba99-49b2-a8fe-2c8efaad2649',
  'filename': 'file-20240815-7UvOUQ-sensitive_document.pdf',
  'bytes': 62430,
  'purpose': 'assistants',
  'created_at': '2024-08-15T02:22:06.716130',
  'object': 'file'
}

from vlmrun.client.types import PredictionResponse

# Submit the document for redaction
response: PredictionResponse = client.document.execute(
    name="healthcare/phi-redaction",
    version="latest",
    file_ids=[response.id],  # Use the file ID from the upload step
    batch=True
)
print(f"Document redaction job submitted:\n {response.model_dump()}")

Document redaction job submitted:
{
  "id": "052cf2a8-2b84-45f5-a385-ccac2aae13bb",
  "created_at": "2024-08-15T02:22:09.157788",
  "response": null,
  "status": "pending"
}

# Wait for the job to complete
response: PredictionResponse = client.predictions.wait(
    id=response.id,
    timeout=120,
)
print(f"Job completed:\n {response.model_dump()}")

​

Industry Use Cases

​

Healthcare - PHI Redaction

​

Other Industries

• Attorney-client privileged information
• Case numbers and court identifiers
• Witness names and contact information
• Settlement amounts and financial details

• Account numbers and routing numbers
• Credit card and payment information
• Social Security Numbers
• Investment account details

• Employee personal information
• Salary and compensation data
• Performance review details
• Background check information

• Property addresses and legal descriptions
• Purchase prices and financial terms
• Personal information of buyers/sellers
• Loan and mortgage details

• Policy numbers and claim identifiers
• Personal health information
• Financial and asset information
• Beneficiary details

​

Example Response

{
  "id": "052cf2a8-2b84-45f5-a385-ccac2aae13bb",
  "created_at": "2024-08-15T02:22:09.157788",
  "status": "completed",
  "response": {
    "redacted_uri": "https://storage.googleapis.com/vlm-userdata/agents/healthcare/phi-redaction/redacted-sensitive_document-20240815-a1b2c3d4.pdf",
    "redacted_items": [
      {
        "phi_type": "name"
      },
      {
        "phi_type": "date_elements"
      },
      {
        "phi_type": "telephone_number"
      },
      {
        "phi_type": "email_address"
      },
      {
        "phi_type": "medical_record_number"
      },
      {
        "phi_type": "geographic_subdivision"
      }
    ]
  }
}

• redacted_uri: A secure, time-limited URL to download the redacted document with all sensitive information visually obscured
• redacted_items: A list of all sensitive information types that were detected and redacted in the document

​

Complete Example

from vlmrun.client import VLMRun
from vlmrun.client.types import PredictionResponse, FileResponse
from pathlib import Path

# Initialize the client
client = VLMRun(api_key="<your-api-key>")

# Upload the document
file_response: FileResponse = client.files.upload(
    file=Path("path/to/sensitive_document.pdf")
)

# Submit for redaction
response: PredictionResponse = client.document.execute(
    name="healthcare/phi-redaction",
    version="latest",
    file_ids=[file_response.id],
    batch=True
)

# Wait for completion
completed_response = client.predictions.wait(response.id, timeout=120)

# Access the redacted document
redacted_uri = completed_response.response["redacted_uri"]
redacted_items = completed_response.response["redacted_items"]

print(f"Redacted document available at: {redacted_uri}")
print(f"Sensitive information types redacted: {[item['phi_type'] for item in redacted_items]}")

​

Using URLs Instead of File Upload

# Process a document from URL
response: PredictionResponse = client.document.execute(
    name="healthcare/phi-redaction",
    version="latest",
    urls=["https://example.com/sensitive_document.pdf"],
    batch=True
)

# Wait for completion
completed_response = client.predictions.wait(response.id, timeout=120)
print(completed_response.response)

​

Batch Processing Multiple Documents

# Upload multiple documents
file_paths = [
    "path/to/document1.pdf",
    "path/to/document2.pdf",
    "path/to/document3.pdf"
]

file_ids = []
for file_path in file_paths:
    file_response = client.files.upload(file=Path(file_path))
    file_ids.append(file_response.id)

# Submit batch redaction job
response: PredictionResponse = client.document.execute(
    name="healthcare/phi-redaction",
    version="latest",
    file_ids=file_ids,
    batch=True
)

# Wait for completion
completed_response = client.predictions.wait(response.id, timeout=300)

# Process results
for redacted_item in completed_response.response["redacted_items"]:
    print(f"Redacted sensitive information type: {redacted_item['phi_type']}")

​

Compliance and Data Security

• 🏥 Healthcare Compliance: Follows HIPAA Safe Harbor method for PHI de-identification
• 🏛️ Legal Compliance: Supports attorney-client privilege and confidentiality requirements
• 💰 Financial Compliance: Meets PCI DSS and financial data protection standards
• 🔒 Secure Processing: All documents are processed in compliant infrastructure with encryption in transit and at rest
• 🔑 Access Controls: Robust authentication and authorization mechanisms protect sensitive data
• 📝 Audit Logging: Comprehensive audit trails for all redaction activities
• ⏰ Secure URLs: Redacted documents are provided via time-limited, secure URLs that automatically expire

​

Use Cases

​

Healthcare

• 📋 Research Data Preparation: Automatically redact PHI from medical records before sharing with research teams
• 🔄 Document Sharing: Safely share patient documents with external providers and insurance companies
• 💾 Data Archival: Prepare historical medical records for long-term storage while removing PHI
• 🏥 Quality Assurance: Create de-identified versions for training and educational purposes

​

Legal

• ⚖️ Discovery Process: Redact privileged information from documents during legal discovery
• 📄 Public Filing: Prepare court documents for public filing by removing sensitive client information
• 🤝 Client Confidentiality: Protect attorney-client privilege in shared documents

​

Financial Services

• 💳 Compliance Reporting: Remove PII from financial reports and audit documents
• 🏦 Data Sharing: Safely share customer data with third-party vendors and partners
• 📊 Analytics: Enable financial analytics while protecting customer privacy

​

Human Resources

• 👥 Employee Records: Redact personal information from HR documents for compliance
• 📈 Performance Analysis: Analyze workforce data while protecting employee privacy
• 🔍 Background Checks: Prepare sanitized versions of background check reports

​

Supported Document Types

• PDF Documents - Reports, contracts, legal briefs, medical records
• Scanned Images - Faxed documents, handwritten forms, ID cards, insurance documents
• Multi-page Documents - Complete case files, comprehensive reports, patient histories
• Mixed Content - Documents containing both text and images, forms with signatures

​

Related Guides

​

Try our Document -> JSON API today