QueryCraft: A Natural Language-Driven NoSQL Database Querying System Powered by Large Language Models

Abstract

QueryCraft is an interactive MongoDB interface for non-relational NoSQL document databases through which users ask natural language questions, inspect generated aggregation pipelines, and execute queries without manually writing MongoDB code. The interface includes a nested schema browser, an inspectable program view, a formatted results panel, and a local history panel for reviewing previous attempts. The text-to-NoSQL generator is replaceable; this implementation uses the Schema-as-Data Grounding (SAG) solver from the Text-to-NoSQL Dataset (TEND) release, with schema grounding, guarded decoding, execution repair, and result consistency. The demo exposes generated pipelines, execution results, runtime errors, and solver metadata, giving users and researchers a concrete view of successful queries and failure cases.