Home/ Part XII — Building Real Products (End-to-End Projects)/36. Project 5: Data-to-Insights Analyst Tool/36.3 Schema inference and validation

36.3 Schema inference and validation

Overview and links for this section of the guide.

On this page

Resolving Ambiguity

If the CSV has columns `sales_jan` and `sales_feb`, and the user asks "Show me sales," the model should ask: "Do you mean January, February, or total?"

This is the Schema Inference step. The model looks at the prompt and the schema, and if the mapping isn't 1:1, it raises a flag.

Validating the Code

Before running the generated Python, scan it.

  • Allowed imports: `pandas`, `numpy`, `matplotlib`.
  • Banned imports: `os`, `sys`, `subprocess`, `requests`.

This is a basic safety layer. The real safety comes from the execution sandbox.

Where to go next