Visualization

Overview

LangExtract's visualization features help you:

• See source grounding: Visual highlights showing where each extracted value came from
• Explore structures: Interactive views of nested extracted data
• Debug extractions: Identify issues and understand model behavior
• Share results: Export visualizations for documentation and presentations

Span Highlighting

One of the most powerful visualization features is span highlighting, which shows exactly where in the source text each extracted value originated:

• Color-coded highlights: Different colors for different field types
• Hover interactions: Hover over highlights to see field names and values
• Multiple spans: Handle cases where a field spans multiple locations
• Overlapping spans: Visual distinction for overlapping extractions

This makes it immediately clear where information came from and helps verify extraction accuracy.

Structured Data View

LangExtract provides formatted views of extracted data structures:

• Formatted JSON: Pretty-printed structured data
• Tree view: Hierarchical display of nested structures
• Field details: Expandable views showing field metadata
• Source links: Click through to source locations

Interactive Exploration

Visualization tools support interactive exploration:

• Click to navigate: Click on extracted values to jump to source locations
• Filter views: Filter by field type, confidence, or other attributes
• Search: Search through extracted values and source text
• Compare: Side-by-side comparison of multiple extractions

Using Visualization in Code

LangExtract provides programmatic access to visualization features:

import langextract as lx

result = lx.extract(
    text_or_documents=text,
    prompt_description="Extract information",
    examples=[...]
)

# Generate visualization
visualization = lx.visualize(result, source_text=text)

# Display in Jupyter notebook
visualization.show()

# Export to HTML
visualization.export_html("extraction.html")

# Export to image
visualization.export_image("extraction.png")

Jupyter Notebook Integration

LangExtract works seamlessly in Jupyter notebooks:

• Inline display: Visualizations render directly in notebook cells
• Interactive widgets: Use interactive controls to explore extractions
• Export options: Save visualizations as part of notebook outputs

This makes LangExtract ideal for exploratory data analysis and documentation.

Web-Based Visualization

For production deployments, LangExtract can generate web-based visualizations:

• HTML export: Standalone HTML files with interactive visualizations
• Embeddable widgets: Embed visualizations in web applications
• API endpoints: Serve visualizations via REST APIs

See the RadExtract demo for a live example of web-based visualization.

Best Practices

• Use for debugging: Visualize extractions to understand model behavior
• Verify accuracy: Check source highlights to verify extraction quality
• Share results: Export visualizations to share with stakeholders
• Document workflows: Include visualizations in documentation and reports