AI-Powered Text Extraction Tool

Extract Structured Data from Text, Documents, and Images

Version 3.0.1 | 9 October 2026 | Updated since the paper. What changed?

Authors: Fadel M. Megahed, Ying-Ju Chen, Yamin Dahwich, Arthur Carvalho, L. Allison Jones-Farmer, Ibrahim Yousif, and Inez M. Zwetsloot

Contact: fmegahed@miamioh.edu

A collaboration between Miami University, the University of Dayton, the University of Cambridge, and the University of Amsterdam

How to Use This App

Quick Demo

Click the 'Load Demo Data' button to load two sample NHTSA recall notices with pre-configured extraction fields. Then click 'Extract Data' to see the model extract structured information from the text. You can also add more fields, change field labels and types, and modify descriptions. (Note: The field description is the instruction the model receives. The label names the column in the results.)

Input Methods

Choose from six input methods to provide your source content. Only the tab that is open when you click 'Extract Data' is used.

Demo Data: Try the app with pre-loaded NHTSA vehicle recall examples.
Paste Text: Directly paste text content. Separate multiple items with double line breaks.
Text File: Upload .txt, .csv (single column), or .md files.
Readable PDF: Upload a machine-readable PDF. Text is extracted automatically.
Scanned PDF: Upload scanned/image-based PDFs. Pages are converted to images for vision processing.
Upload Images: Upload images (PNG, JPEG, WebP, GIF) for vision-based extraction.

Notes:

Readable PDF and Scanned PDF are limited to the first 5 pages, and Upload Images to 5 images, in this free app.
Any uploaded file must be at most 5 MB, to keep the app responsive and avoid upload timeouts.
The '?' buttons next to each part of the app explain what it does and what is sent to the model.

Custom Fields

Customize your extraction by adjusting:

The number of fields to extract (up to 10 )
Your own field labels (the column names of the results)
Detailed field descriptions to guide the model
The field type (text, number, yes/no, or list of text)

You can start from a template for a common quality document (recall notice, 8D / corrective action report, customer complaint record, inspection or nonconformance note), and save your field definitions to a file to reuse them later.

Specific field descriptions give the model less room for interpretation. Always check the extracted values against the source documents.

Structured Data Extracted Using AI

Reproduce in Code

The code below runs the same extraction, with your current field definitions and the model gpt-6-luna , on your own computer and your own data. It needs your own OpenAI API key.

R code (ellmer package)

Python code (chatlas package)

Tips for Better Results

Use specific field descriptions and say what to return when the information is absent
Start with more fields and remove unnecessary ones later
If results are inaccurate, try rephrasing your field descriptions
For multiple text blocks , separate each with a double line break
Each text block should contain complete information for all fields you want to extract
Check the results against the source documents before you use them

Technical Note

This app sends its requests to the OpenAI model gpt-6-luna through chat_openai() from the ellmer package. The app does not provide an option to change the model.

For complete privacy , consider using local open-weight models via chat_ollama() from ellmer, which connects to Ollama for running models locally on your machine.

ellmer also supports models from other providers. For example, Claude models require chat_anthropic() and Gemini models require chat_google_gemini() , each with their respective API keys configured in the R environment. The code under 'Reproduce in Code' can be adapted by replacing the chat_openai() call.