ask command

[jqnatividad]

Since we have the schema, frequency and summary statistics about a CSV, we should be able to deploy a model that can formulate a SQL query based on a natural language question.

And with the sqlp command, we can run the generated SQL query to answer the question with confidence without having to fear hallucinations nor the model ingesting sensitive data to train on. It just trains on the "data dictionary" - the schema, summary statistics and frequency table describing the dataset.

This will be the foundation of our "Answering People Interface".