GraphGPT converts unstructured text into knowledge graphs, ideal for researchers and power users exploring data relationships. It leverages GPT-3 to extract entities and relationships, visualizing them for easier understanding and allowing iterative graph updates.

How It Works

The project uses a few-shot prompt to guide GPT-3 in understanding and formatting natural language into a JSON structure representing entities and their relationships. This JSON is then rendered as a graph visualization. The approach allows for successive queries to update the existing graph or create new structures, offering a dynamic way to build knowledge representations.

Quick Start & Requirements

• Install dependencies: npm install
• Run the application: npm run start
• Requires an OpenAI API key.
• Dependencies: react-graph-vis.
• Setup involves cloning the repo and running npm commands.

Highlighted Details

• Converts natural language text (movie synopses, Wikipedia passages, video transcripts) into knowledge graphs.
• Supports successive queries to update existing graph states or create new structures.
• Graph visualization allows for entity and relationship mapping.

Maintenance & Community

This is noted as a toy project built over a weekend. For production use cases, the author recommends checking out GPT Index.

Licensing & Compatibility

The README does not specify a license.

Limitations & Caveats

Significant latency issues exist due to OpenAI API call times, with responses taking up to 20 seconds. The project is described as a "toy project" and may not be suitable for production environments.