Unstructured Text¶

pip install structx-llm

# Complete document processing support (recommended)
pip install structx-llm[docs]

# Individual components
pip install structx-llm[pdf]     # PDF conversion and processing
pip install structx-llm[docx]    # DOCX support via docling

# Process a DOCX contract by converting it to PDF for optimal extraction
result = extractor.extract(
    data="scripts/example_input/free-consultancy-agreement.docx", # Converted to PDF -> multimodal
    query="extract the parties, effective date, and payment terms"
)

# PDFs are processed directly with multimodal support
result = extractor.extract(
    data="scripts/example_input/S0305SampleInvoice.pdf",          # Direct multimodal processing
    query="extract the invoice number, total amount, and line items"
)

# Extract from a markdown file (converted to styled PDF)
result = extractor.extract(
    data="README.md",
    query="extract installation instructions and requirements"
)

# Extract from log files (converted to PDF with monospace formatting)
result = extractor.extract(
    data="system.log",
    query="extract error events and timestamps"
)

# Extract from code files (converted to PDF with syntax highlighting)
result = extractor.extract(
    data="script.py",
    query="extract function names and their purposes"
)

# Extract from PDF (direct multimodal processing - no conversion needed)
result = extractor.extract(
    data="contract.pdf",
    query="extract parties, dates, and payment terms"
)

# Works with complex PDFs including tables, images, and layouts
result = extractor.extract(
    data="financial_report.pdf",
    query="extract financial metrics and performance indicators"
)

# Extract from DOCX (docling → markdown → styled PDF → multimodal)
result = extractor.extract(
    data="project_proposal.docx",
    query="extract project timeline, budget, and deliverables"
)

# Preserves document structure including tables and formatting
result = extractor.extract(
    data="meeting_notes.docx",
    query="extract action items and responsible parties"
)

text = """
Incident Report #2024-001
Date: January 15, 2024
System: Production Server-01
Issue: High CPU usage detected at 92%
Resolution: Memory leak patched, monitoring increased
"""

result = extractor.extract(
    data=text,
    query="extract incident details including ID, date, system, and resolution"
)

# Default: multimodal PDF processing (recommended)
result = extractor.extract(
    data="document.docx",
    query="extract key information"
    # mode="multimodal_pdf" is default
    # use_multimodal=True is default
)

result = extractor.extract(
    data="document.txt",
    query="extract key information",
    mode="simple_text",           # Override default
    chunk_size=1500,             # Custom chunk size
    chunk_overlap=300            # Custom overlap
)

result = extractor.extract(
    data="document.pdf",
    query="extract key information",
    mode="simple_pdf",           # Basic PDF text extraction
    chunk_size=2000,
    chunk_overlap=200
)

result = extractor.extract(
    data="document.csv",
    query="extract key information",
    file_options={
        "encoding": "utf-8",     # For text files
        "sep": ";",              # For CSV files
        "sheet_name": "Data"     # For Excel files
    }
)

# For very large documents, consider processing in smaller sections
# or using simple modes if memory is constrained
result = extractor.extract(
    data="huge_document.pdf",
    query="extract specific section data",
    # Consider simple_pdf mode for memory-constrained environments
)

# Process multiple documents individually for best results
documents = ["doc1.pdf", "doc2.docx", "doc3.txt"]

results = []
for doc in documents:
    result = extractor.extract(
        data=doc,
        query="extract key information"
    )
    results.append(result)

# Take advantage of full document context
result = extractor.extract(
    data="contract.pdf",
    query="""Extract all contract details including:
    - Party names and addresses
    - Contract dates (start, end, signature)
    - Payment terms and amounts
    - Key obligations and deliverables
    - Termination clauses"""
)

# Extract comprehensive contract information from a PDF
result = extractor.extract(
    data="service_agreement.pdf",
    query="""Extract the following contract details:
    - Contracting parties (names, roles, addresses)
    - Contract effective date and duration
    - Service descriptions and deliverables
    - Payment terms, amounts, and schedules
    - Key obligations for each party
    - Termination conditions and notice periods
    - Governing law and dispute resolution"""
)

# Access extracted data
for contract in result.data:
    print(f"Contract between: {contract.parties}")
    print(f"Effective: {contract.effective_date}")
    print(f"Services: {contract.services}")

# Extract API documentation from markdown files
result = extractor.extract(
    data="api_docs.md",
    query="""Extract API endpoint information:
    - Endpoint URLs and HTTP methods
    - Required and optional parameters
    - Request/response examples
    - Authentication requirements
    - Rate limiting information
    - Error codes and descriptions"""
)

# Process quarterly reports from DOCX files
result = extractor.extract(
    data="Q4_financial_report.docx",
    query="""Extract financial metrics and insights:
    - Revenue figures by quarter and year-over-year
    - Profit margins and operating expenses
    - Key performance indicators (KPIs)
    - Market outlook and forward guidance
    - Risk factors and mitigation strategies
    - Executive summary highlights"""
)

# Analyze system logs for incidents
result = extractor.extract(
    data="system_logs.log",
    query="""Extract system incidents and events:
    - Error events with timestamps and severity
    - System performance metrics and alerts
    - User activities and authentication events
    - Network connectivity issues
    - Hardware or software failures
    - Recovery actions and resolutions"""
)

# If conversion fails, the library automatically falls back to simple text
# You can also force simple text mode:
result = extractor.extract(
    data="problematic_document.docx",
    query="extract information",
    mode="simple_text"  # Skip PDF conversion
)

# Install complete document processing support
pip install structx-llm[docs]

# Or install specific components
pip install weasyprint markdown docling PyPDF2

# Use simple processing for very large files
result = extractor.extract(
    data="large_document.pdf",
    query="extract key information",
    mode="simple_pdf",     # Reduce memory usage
    chunk_size=1500,       # Smaller chunks
    chunk_overlap=150      # Reduced overlap
)

# If docling conversion fails, fallback is automatic
# You can check what processing method was used:

# Check the DataFrame structure after reading
df = extractor.extract(data="document.docx", query="test")
print(df.columns)  # Will show processing method used

import logging
from loguru import logger

# Enable debug logging
logger.add("extraction_debug.log", level="DEBUG")

result = extractor.extract(
    data="problematic_document.pdf",
    query="extract information"
)

import time

start_time = time.time()
result = extractor.extract(
    data="document.pdf",
    query="extract information"
)
processing_time = time.time() - start_time

print(f"Processing time: {processing_time:.2f} seconds")
print(f"Success rate: {result.success_rate:.1f}%")
print(f"Items extracted: {result.success_count}")