GraphGPT

There are three significant issues related to subgraph interactions:

1. Accessing and querying relevant on-chain data is complex and time-consuming.
2. Extracting insights from data requires additional steps and technical skills.
3. Turning raw data into clear visuals is always challenging for users.

GraphGPT aims to make interacting with subgraphs easier by:

1. Providing a user-friendly interface for querying data.
2. Leveraging AI to assist in extracting insights from data.
3. Simplifying data visualization to improve analysis efficiency.

How it works?

Step1. The user inputs the query through the "LLM planner" to perform keyword extraction, generating an easy-read query JSON file. Step2. The JSON file is sent to "LLM GraphQL Intelligence," which produces a GraphQL Query and then calls the subgraph API to output the graphQL query result JSON file. Step3. The graphQL query result JSON file is then sent to the "LLM Data Analysis," which generates user visualization and insights.

We are focused on building an AI agent system that can handle various user queries when interacting with The Graph. We aim to provide a seamless user experience and visualize data for better understanding.

Technologies used:

The Graph: By leveraging The Graph, we can query data from subgraphs such as Uniswap, Lido, Snapshot, and more. AI Agent: We use AI LLM and AI Agent techniques to generate GraphQL queries for querying subgraphs. Visualization and Postprocessing: OpenAI LLM is employed to generate plotting code and save the script for data visualization. Frontend: Python Streamlit - chatbox Backend: The Graph, Python, LLM AI