data copilot for self-service analytics

[wgong]

Based on Vanna and streamlit, I prototyped a small app called data copilot for self-service analytics, would like to share it here.

• A 5 min demo video at YouTube
• Source codes available at GitHub

Your feedback would be appreciated!

2 questions to Vanna maintainers:

1. When I work with multiple data sources, should I use separate vector-stores to avoid schema context mixing?
2. If I like to use single vector-store for multiple data-sources, how should I extend training data schema by adding data-source metadata (e.g. Database type and connection URL), similar to the name column in chroma collections table storing training data type like ddl/sql/document?
   Thanks