Smart Artifact Parser

Extract structured medical information from documents using AI.

Smart Artifact Parser uses Docling for document parsing and Claude for intelligent extraction, converting unstructured medical documents into clean, validated JSON.

Features

• Multi-format support: PDF, DOCX, TXT, and images (PNG, JPG, TIFF, BMP)
• OCR included: Automatic text extraction from scanned documents and images
• Structured output: Validated JSON with Pydantic schemas
• Extensible: Add new extraction fields by modifying the schema
• CLI interface: Simple command-line tool for batch processing

Installation

Requires Python 3.12+ and uv.

# Clone the repository
git clone https://github.com/danjamk/smart-artifact-parser.git
cd smart-artifact-parser

# Install dependencies
uv sync

Configuration

Create a .env file with your Anthropic API key:

cp .env.example .env

Edit .env and add your key:

ANTHROPIC_API_KEY=your_api_key_here

Get an API key at console.anthropic.com.

Usage

# Extract from a document
uv run python -m src.cli document.pdf

# Specify output directory
uv run python -m src.cli document.pdf --output-dir ./results

# Using the installed command (after uv sync)
uv run smart-parser document.pdf

Output

Results are saved as JSON files with timestamped names:

output/document_20250126_143022.json

Extracted Fields

The parser extracts the following information:

Field	Description
document_type	visit_note, lab_result, discharge_summary, prescription, referral, imaging_report, or other
document_date	Date of the document/visit
provider	Healthcare provider name, specialty, and facility
chief_complaint	Primary reason for the visit
assessment	Provider's clinical assessment
diagnoses	List of diagnoses with optional ICD-10 codes
medications	Medications with dosage, frequency, and instructions
follow_up_instructions	Follow-up care instructions

Example Output

{
  "document_type": "visit_note",
  "document_date": "2025-01-15",
  "provider": {
    "name": "Dr. Jane Smith",
    "specialty": "Internal Medicine",
    "facility": "City Medical Center"
  },
  "chief_complaint": "Persistent cough for 2 weeks",
  "assessment": "Acute bronchitis, improving",
  "diagnoses": [
    {
      "description": "Acute bronchitis",
      "icd_code": "J20.9"
    }
  ],
  "medications": [
    {
      "name": "Amoxicillin",
      "dosage": "500mg",
      "frequency": "Three times daily",
      "instructions": "Take with food for 7 days"
    }
  ],
  "follow_up_instructions": "Return if symptoms worsen or fever develops",
  "source_file": "visit_note.pdf",
  "extracted_at": "2025-01-26T14:30:22.123456"
}

Supported Document Types

Format	Extension	Notes
PDF	.pdf	Native text and scanned (via OCR)
Word	.docx	Microsoft Word documents
Text	.txt	Plain text files
Images	.png, .jpg, .jpeg, .tiff, .bmp	OCR extraction

Extending the Schema

To extract additional fields, modify src/schemas.py:

1. Add new Pydantic models for complex types
2. Add fields to MedicalDocumentExtraction
3. Claude will automatically extract the new fields

Example - adding patient vitals:

class Vitals(BaseModel):
    blood_pressure: str | None = Field(default=None, description="Blood pressure reading")
    heart_rate: int | None = Field(default=None, description="Heart rate in BPM")
    temperature: float | None = Field(default=None, description="Temperature in Fahrenheit")

class MedicalDocumentExtraction(BaseModel):
    # ... existing fields ...
    vitals: Vitals | None = Field(default=None, description="Patient vital signs")

Project Structure

smart-artifact-parser/
├── src/
│   ├── cli.py          # Typer CLI entry point
│   ├── parser.py       # Document parsing (Docling)
│   ├── extractor.py    # Claude API extraction
│   └── schemas.py      # Pydantic data models
├── output/             # Default output directory
├── pyproject.toml      # Project configuration
└── .env.example        # Environment template

How It Works

1. Parse: Docling converts the document to markdown, preserving tables and structure
2. Extract: Claude analyzes the text and extracts structured data using tool_use
3. Validate: Pydantic validates the extracted data against the schema
4. Output: Results saved as JSON with metadata

Requirements

• Python 3.12+
• uv package manager
• Anthropic API key

Dependencies

• docling - Document parsing
• anthropic - Claude API client
• pydantic - Data validation
• typer - CLI framework

License

MIT License - see LICENSE for details.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.