Data Agent is a project designed to utilize large models to generate SQL queries for fetching data, subsequently creating corresponding reports and charts. The project architecture includes:

• auto_plot_mat.py: Script for generating charts, which can deal with most of the situations of the dataframe with time_column, numeric_column and category_column.
• main.py: Main function script.
• data/: Directory containing two example CSV files for testing.

1. Upload Data: Begin by uploading one or more CSV files. These files will be used for generating SQL queries and subsequent analysis.

2. Explore Data Schema: : Upon uploading data, the system generates a description of the schema, allowing users to understand the structure of their data for better insights and query formulation.

3. Ask Questions: Describe the data you want to query in the provided textbox.

4. View Reports and Visualizations: The system will generate SQL queries based on your description, fetch the data, debug the sql and regenerate until there is no errors, create visualizations, and present a detailed report.

• Data Schema Description:: Upon data upload, the system provides a detailed description of the schema, including data types and descriptions of columns. This enables users to understand their data better and formulate more precise queries.

• Intelligent SQL Querying: Utilizes a large language model to generate complex SQL queries based on natural language descriptions.

• Error Correction: In case of SQL errors, the llm can automatically debug the sql and regenerate the correct SQL until there is no errors, allowing for a smoother workflow.

• Data Visualization: The system automatically generates visualizations such as charts based on the fetched data.

• Workflow: This project is designed with a workflow in mind, ensuring a structured approach to data analysis:

  1. Data Upload: CSV files are uploaded for processing.

  2. Data Schema Generation: The system generates a schema description to help users understand the data structure.

  3. Query Description: Users describe the data they want to analyze in natural language.

  4. SQL Query Generation: The system generates SQL queries based on the description.

  5. SQL Execution: Queries are executed on the uploaded data to fetch relevant information.

  6. Report and Visualization: The system generates detailed reports and visualizations from the fetched data.

  7. Error Handling: If there are SQL errors, the system provides suggestions for correction.

  8. Final Report: Users receive a comprehensive report, including SQL queries used, retrieved data, and visualizations.

Here are a few examples of the types of questions you can ask:

• "What is the average income of individuals with a bachelor's degree?"
• "What is the average income by geographic region?"
• "How many individuals are single?"

1. Ensure you have the necessary dependencies installed.

2. Run main.py with your OpenAI API key as an argument. Example:

   python main.py --openai_key YOUR_OPENAI_API_KEY
   

3. Follow the instructions in the Gradio interface to upload data, ask questions, and view reports.

• openai: OpenAI Python SDK for interfacing with GPT-3.5 Turbo.
• pandas: For data manipulation and analysis.
• pandasql: For running SQL queries on DataFrame objects.
• gradio: For creating the user interface.

• SQL Error Correction: If there are errors in the generated SQL queries, the system guides the user by suggesting corrected queries, ensuring a smoother workflow.

For any inquiries or feedback, please contact Dalin Wang at [email protected].

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