CatLLM

A Python package designed for researchers conducting text, image, and PDF analysis using large language and vision models, providing systematized access to LLM capabilities for extracting categorical information from unstructured data. The package delivers output in a consistent, highly reproducible format optimized for statistical analysis, with optional CSV export functionality for seamless integration into research workflows.

Core Functionality:

Text Analysis Capabilities:

• Extract multiple categories present within individual text responses

• Identify the single most prominent category within individual text responses

• Determine the most frequently occurring categories across entire text corpora

Image Analysis Capabilities:

• Classify multiple categorical features present within individual images

• Conduct open-ended feature identification within individual images

• Generate quality scores for individual images using reference image comparisons

PDF Analysis Capabilities:

• Classify multiple categories present within individual PDF pages
• Process PDFs using image, text, or combined analysis modes
• Extract structured data from documents with charts, tables, and figures

Research-Oriented Design:
The package standardizes LLM interactions to ensure consistent output formatting across different models and analysis tasks. All functions return structured data objects ready for immediate statistical analysis, eliminating the need for extensive post-processing. Users can export results directly to CSV format for integration with statistical software packages, streamlining the transition from qualitative data collection to quantitative analysis. This design approach addresses common reproducibility challenges in LLM-assisted research by providing standardized interfaces and consistent output schemas regardless of the underlying model or complexity of the analytical task.