RTS+: Reliable Text to SQL
Kaiwen Chen, Yueting Chen, Nick Koudas, Xiaohui Yu · 2025
The reliable translation of natural language queries to SQL is piv- otal for empowering non-technical users to interact with databases. Current state-of-the-art systems fail to holistically address ambi- guities or schema-linking challenges and may generate erroneous SQL leading to faulty query generation. We present RTS+ (Reliable Text-to-SQL Plus), a novel interactive system that builds on the con- cepts of Reliable Text-to-SQL (RTS) [3 ]. RTS+ focuses on improving user experience by incorporating adaptive abstention mechanisms and interactive feedback loops during query generation. Through this demo, we illustrate RTS+'s capabilities using a rich graphical interface that guides users through schema-linking decisions, query generation, verification, and refinement. Users can explore RTS+ on real-world datasets with detailed analytics for understanding abstention-triggered interventions and their outcomes. Using this system, users can achieve perfect scores at popular benchmarks correctly answering a few questions, if required.