Overview
Most people can't write SQL. This playbook puts a natural-language layer over curated open datasets so staff ask questions in plain Arabic/English and get answers with the source rows.
How it works
- Catalogue the datasets and their columns with clear descriptions.
- AI translates a plain-language question into a safe, read-only query.
- Return the answer with the underlying rows and the query used (for trust).
- Log unanswered questions to improve the catalogue.
Guardrails
Read-only, scoped to curated datasets, with the generated query always shown.