Extralit

Extract structured data from scientific literature with human validation

Extralit is an open-source platform that transforms how researchers extract structured data from scientific literature. Want to get started? Check out our documentation.

Why use Extralit?

Accelerate Scientific Data Collection

Manual data extraction from research papers is slow and error-prone, often taking 6-12 months for systematic reviews. Extralit combines AI-powered extraction with human validation to reduce this to weeks while maintaining research-grade accuracy.

Take Control of Your Research Data

Most scientific data extraction tools are inflexible black boxes. Extralit is different - it's open source and puts you in control. Define custom extraction schemas, validate results, and integrate with your existing research workflows.

Scale Your Literature Reviews

Whether you're conducting a systematic review, meta-analysis, or building a scientific knowledge base, Extralit helps you efficiently process hundreds of papers. Our platform handles complex tables, figures, and relationships while preserving scientific rigor.

🏘️ Community

We're an open-source project built for researchers, by researchers. Here's how to get involved:

• Slack Community: Connect with other researchers and developers
• Documentation: Learn how to use and contribute to Extralit
• Roadmap: See what we're building and share your ideas

Real-World Impact

Extralit is already accelerating research at leading institutions:

• Gates Foundation: Reduced systematic review time for malaria intervention studies from 6 months to 6 weeks
• Life Science Research: Streamlined extraction of clinical trial endpoints, genetic markers, and intervention protocols
• Meta-Analysis: Enabled rapid synthesis of evidence across hundreds of papers while maintaining rigorous validation

👨‍💻 Getting Started

Installation

Install Extralit using pip:

pip install extralit

Initialize the client:

import extralit as ex

client = ex.Extralit(
    api_url="https://your-deployment-url",
    api_key="your-api-key"
)

Create an extraction schema

Define the fields the dataset will display and the questions annotators answer:

settings = ex.Settings(
    fields=[
        ex.TextField(name="text", title="Document Text"),
        ex.ImageField(name="image", required=False),
    ],
    questions=[
        ex.LabelQuestion(name="label", labels=["positive", "negative"]),
        ex.TextQuestion(name="comment", use_markdown=False),
    ],
    guidelines="Classify the sentiment and provide a comment.",
)

Add documents and start extraction

Create a dataset with the schema, then log records. Pre-filled values (e.g. from an LLM) become suggestions that annotators can accept or override.

dataset = ex.Dataset(name="my_extraction_dataset", settings=settings).create()

records = [
    {"text": "This product is amazing!", "label": "positive"},
    {"text": "Terrible experience.", "label": "negative"},
]
dataset.records.log(records)

for record in dataset.records(with_suggestions=True):
    print(record.id, record.fields)

Need more help? Check out our detailed tutorials.

🥇 Contributors

Want to contribute? Great! Check out our contribution guide or join our Slack community.