• Data Copilot Demo:

Data Copilot: My goal is to create an assistant that helps me quickly access to product data. Why? There are two reasons:

• 1. I need rapid, anytime access to 360-degree information of feature performance & operation (I am the Product Manger)
• 2. I have worked directly with ChatGPT and encountered several limitations:
  • ChatGPT does not allow me to handle Google Sheets directly
  • Uploading and processing data with ChatGPT is a costly (tokens), time-consuming process, and sometimes the risk of having to share data is inevitable --> I need a cheaper, more secure solution

During my research, I discovered PandasAI (a fantastic library, which I recommend to anyone with similar goals) and to help you quickly get started, I created a demo project on Streamlit. You can freely download and explore it.

• How to Combine PandasAI + ChatGPT + Streamlit
  • Streamlit: allows you to build chat interfaces very quickly
  • Basic techniques with PandasAI
    • PandasAI only sends metadata instead of the full data set -> no worries about data leak risks
    • How to handle the response and display it on Streamlit
• How to use ChatGPT and PlanUML to draw ER Diagrams (I use it to check if ChatGPT really understands the relationships between data tables, as a misunderstanding could lead to incorrect processing)
• How to work with Google Sheets

You have to create two environment variables to store the OPENAI_API_KEY and the Credentials information when connecting to Google Sheets before running

export OPENAI_API_KEY="sk-iOZXf..................gqObIX5cd8Z"
export GS_ACCOUNT_JSON='{"type":"service_account","project_id":"...........,"universe_domain":"googleapis.com"}'

• Update after I have more time and further develop my idea