docling-analysis-framework 1.0.1

AI-ready analysis framework for PDF and Office documents using Docling

Docling Analysis Framework

AI-ready analysis framework for PDF and Office documents using Docling for content extraction. Transforms Docling's output into optimized chunks and structured analysis for AI/ML pipelines.

📝 For text, code, and configuration files, use our companion document-analysis-framework which uses only Python standard library.

🚀 Quick Start

Simple API - Get Started in Seconds

import docling_analysis_framework as daf

# 🎯 One-line analysis with Docling extraction
result = daf.analyze("document.pdf")
print(f"Document type: {result['document_type'].type_name}")
print(f"Pages: {result['document_type'].pages}")
print(f"Handler used: {result['handler_used']}")

# ✂️ Smart chunking for AI/ML
chunks = daf.chunk("document.pdf", strategy="auto")
print(f"Created {len(chunks)} optimized chunks")

# 🚀 Enhanced analysis with both analysis and chunking
enhanced = daf.analyze_enhanced("document.pdf", chunking_strategy="structural")
print(f"Document: {enhanced['analysis']['document_type'].type_name}")
print(f"Chunks: {len(enhanced['chunks'])}")

# 💾 Save chunks to JSON
daf.save_chunks_to_json(chunks, "chunks_output.json")

# 💾 Save analysis to JSON  
daf.save_analysis_to_json(result, "analysis_output.json")

Advanced Usage

import docling_analysis_framework as daf

# Enhanced analysis with full results
enhanced = daf.analyze_enhanced("research_paper.pdf")

print(f"Type: {enhanced['analysis']['document_type'].type_name}")
print(f"Confidence: {enhanced['analysis']['confidence']:.2f}")
print(f"AI use cases: {len(enhanced['analysis']['analysis'].ai_use_cases)}")
if enhanced['analysis']['analysis'].quality_metrics:
    for metric, score in enhanced['analysis']['analysis'].quality_metrics.items():
        print(f"{metric}: {score:.2f}")

# Different chunking strategies
hierarchical_chunks = daf.chunk("document.pdf", strategy="structural")
table_aware_chunks = daf.chunk("document.pdf", strategy="table_aware") 
page_aware_chunks = daf.chunk("document.pdf", strategy="page_aware")

# Process chunks
for chunk in hierarchical_chunks:
    print(f"Chunk {chunk.chunk_id}: {chunk.token_count} tokens")
    print(f"Type: {chunk.chunk_type}")
    # Send to your AI model...

# 💾 Save different strategies to separate files
strategies = {
    "structural": hierarchical_chunks,
    "table_aware": table_aware_chunks,
    "page_aware": page_aware_chunks
}

for strategy_name, chunks in strategies.items():
    daf.save_chunks_to_json(chunks, f"chunks_{strategy_name}.json")
    print(f"Saved {len(chunks)} chunks to chunks_{strategy_name}.json")

Expert Usage - Direct Class Access

# For advanced customization, use the classes directly
from docling_analysis_framework import DoclingAnalyzer, DoclingChunkingOrchestrator, ChunkingConfig

analyzer = DoclingAnalyzer(max_file_size_mb=100)
result = analyzer.analyze_document("document.pdf")

# Custom chunking with config
config = ChunkingConfig(
    max_chunk_size=1500,
    min_chunk_size=200,
    overlap_size=100,
    preserve_structure=True
)

orchestrator = DoclingChunkingOrchestrator(config=config)
# Note: In advanced usage, you'd pass the actual docling_result object

📋 Supported Document Types

Format	Extensions	Confidence	Powered By
📄 PDF Documents	.pdf	95%	Docling extraction
📝 Word Documents	.docx	95%	Docling extraction
📊 Spreadsheets	.xlsx	70%	Docling extraction
📅 Presentations	.pptx	70%	Docling extraction
🖼️ Images with Text	.png, .jpg, .tiff	70%	Docling OCR

🎯 Key Features

🔍 Docling-Powered Extraction

• PDF text extraction - High-quality content extraction
• Table detection - Preserves table structure in markdown
• Figure references - Maintains image/figure relationships
• Header hierarchy - Document structure preservation

🤖 AI Preparation Layer

• Quality assessment - Extraction quality scoring
• Structure analysis - Document type detection and analysis
• Chunk optimization - AI-ready segmentation strategies
• Rich metadata - Page counts, figures, tables, quality metrics

⚡ Smart Chunking Strategies

• Structural chunking - Respects document hierarchy (headers, sections)
• Table-aware chunking - Separates tables from text content
• Page-aware chunking - Considers original page boundaries
• Auto-selection - Document-type-aware strategy selection

📊 Extraction Quality Analysis

• Text coverage - How much text was successfully extracted
• Structure preservation - Whether headers, lists, tables are maintained
• Overall confidence - Combined quality score for AI processing

🎉 Framework Status - COMPLETED

The Docling Analysis Framework is now fully functional and follows the same successful patterns as the XML Analysis Framework:

✅ Completed Features

• 🎯 Simple API: One-line functions for analyze(), chunk(), analyze_enhanced()
• 🔧 Advanced API: Direct class access for customization with DoclingAnalyzer, DoclingChunkingOrchestrator
• ⚙️ Configurable Chunking: ChunkingConfig class for fine-tuning chunk parameters
• 📦 Multiple Strategies: Structural, table-aware, page-aware, and auto-selection chunking
• 💾 JSON Export: Easy export of analysis results and chunks to JSON files
• 🛡️ Enhanced Error Handling: Comprehensive logging and error reporting
• 📊 Quality Metrics: Extraction quality assessment and content analysis
• 🧪 Testing Framework: Complete Jupyter notebook for validation
• 📚 Documentation: Comprehensive README with examples and usage patterns

🚀 Ready for AI/ML Integration

The framework provides everything needed for AI/ML pipelines:

• Token-optimized chunks sized for LLM context windows
• Rich metadata with document structure and quality metrics
• JSON export for vector database ingestion
• Multiple chunking strategies for different document types
• Quality assessment to determine AI suitability

🔄 Comparison with XML Framework

Feature	XML Framework	Docling Framework	Status
Simple API	✅	✅	Complete
Advanced Classes	✅	✅	Complete
Multiple Strategies	✅	✅	Complete
Configuration	✅	✅	Complete
JSON Export	✅	✅	Complete
Error Handling	✅	✅	Complete
Testing Notebooks	✅	✅	Complete
Quality Metrics	✅	✅	Complete

Both frameworks now provide identical API patterns and functionality!

🔧 Installation

# Install Docling first
pip install docling

# Install framework
git clone https://github.com/redhat-ai-americas/docling-analysis-framework.git
cd docling-analysis-framework  
pip install -e .

📖 Usage Examples

Document Analysis with Quality Assessment

from core.analyzer import DoclingAnalyzer

analyzer = DoclingAnalyzer()
result = analyzer.analyze_document("contract.pdf")

# Access extraction quality
quality = result['analysis'].extraction_quality
print(f"Text coverage: {quality['text_coverage']:.2f}")
print(f"Structure score: {quality['structure_score']:.2f}")
print(f"Overall quality: {quality['overall_score']:.2f}")

# Access document insights
findings = result['analysis'].key_findings
print(f"Pages: {findings['page_count']}")
print(f"Tables: {findings['table_rows']}")
print(f"Figures: {findings['figure_count']}")

Advanced Chunking for Academic Papers

# Perfect for research papers with complex structure
chunks = orchestrator.chunk_document(
    "paper.pdf",
    markdown_content,
    docling_result,
    strategy='structural'
)

# Chunks respect paper structure
for chunk in chunks:
    if chunk.chunk_type == 'section':
        print(f"Section: {chunk.metadata['section_title']}")
    elif chunk.chunk_type == 'table':
        print(f"Table data: {len(chunk.content)} chars")
    elif chunk.chunk_type == 'figure':
        print(f"Figure reference preserved")

File Size Management

from core.analyzer import DoclingAnalyzer

# Large PDF processing with limits
analyzer = DoclingAnalyzer(max_file_size_mb=100.0)

result = analyzer.analyze_document("large_manual.pdf")
if 'error' in result:
    print(f"File too large: {result['error']}")
else:
    print(f"Successfully processed {result['file_size']} bytes")

🧪 Framework Ecosystem

This framework is part of a larger document analysis ecosystem:

• xml-analysis-framework - Specialized XML document analysis
• docling-analysis-framework - PDF and Office documents (this package)
• document-analysis-framework - Text, code, and configuration files
• unified-analysis-orchestrator - Routes documents to appropriate frameworks

🔗 Integration with Docling

This framework acts as an AI preparation layer on top of Docling:

1. Docling handles the heavy lifting of document extraction
2. Our framework adds AI-specific analysis and chunking
3. Result is AI-ready structured data and optimized chunks

# What Docling provides:
docling_result = docling.convert("document.pdf")
markdown_content = docling_result.document.export_to_markdown()

# What we add:
ai_analysis = our_analyzer.analyze(docling_result)
ai_chunks = our_chunker.chunk(markdown_content, ai_analysis)
quality_scores = our_quality_assessor.assess(docling_result)

📄 License

MIT License - see LICENSE file for details.

🙏 Acknowledgments

• Powered by Docling for document extraction
• Built for modern AI/ML development workflows
• Part of the AI Building Blocks initiative