Glossary · AI & ML

Few-Shot SQL Prompting

Few-shot SQL prompting is a technique where example question-and-SQL pairs are included in the AI's prompt to teach it the query patterns and conventions of a specific database.

Language models generalize from examples remarkably well. Show a model three examples of questions about your database alongside the correct SQL for each, and its accuracy on the fourth question improves noticeably–it learns your naming conventions, your preferred join paths, and quirks like "active users means status = 2, not status = 'active'". This matters because every database encodes business logic in ways no general model could know. Few-shot prompting is the cheapest way to transfer that knowledge: no model training, just well-chosen examples in the prompt.

Fig — every answer ships with the tables, rows and SQL behind it.