ELTEX: Efficient LLM Token Extraction for Synthetic Data Generation

ELTEX is an experimental tool for domain-driven synthetic data generation using large language models (LLMs). This repository contains a Google Sheets Add-on that allows non-technical users to generate high-fidelity, domain-specific data directly from a familiar spreadsheet interface.

⚠️ This project is currently in active development and internal testing. The add-on is not yet available via the Google Workspace Marketplace.

✨ Features

• Token extraction for guiding synthetic data generation
• Multi-model support (e.g. GPT-4o, Claude 3.7, Gemini 2.0 Pro)
• Seamless integration with Google Sheets

🔧 How to Use (Development Version)

To test the add-on locally, you'll need to install it manually via Google Apps Script.

1. Open Google Apps Script

• Create a new project
• Copy the contents of the src/ folder into your project

2. Configure Project Properties

• Add required API keys via File > Project properties > Script properties
• Required keys:
  • OPENAI_API_KEY
  • ANTHROPIC_API_KEY
  • GOOGLE_API_KEY
• Deduplication service (user-provided):
  • ELTEX does not include a built-in deduplication backend. You can deploy your own deduplication service (for example, using the reference Cloudflare Worker) or any compatible HTTP API.
  • If you use the reference Cloudflare Worker implementation, set the following Script Properties in your ELTEX add-on:
    • DEDUP_API_URL — the URL of your deployed deduplication service
    • DEDUP_API_KEY — your API key for your deduplication service
  • If you implement your own deduplication service, the required properties and their names may differ depending on your API.

If you do not configure a deduplication service, deduplication features will not be available.

3. Deploy as a Test Add-on

• In the Deploy menu, select Test deployments
• Choose Add-on > Editor add-on
• Select your test Google Sheet as the deployment target

📌 Status

• Current phase: Active development
• Marketplace release: Planned
• arXiv paper: "ELTEX: A Framework for Domain-Driven Synthetic Data Generation"

📄 License

This project is licensed under the MIT License. See LICENSE for details.

🙋‍♀️ Contact & Contributions

If you're interested in collaborating, providing feedback, or trying out ELTEX for your organization, feel free to open an issue or reach out directly.