DIY: Helping People Assess the Correctness of Natural Language to SQL Systems
summary
Built a sandbox called DIY that generates a sample database (five records per table), splits a generated SQL into steps, and shows mappings plus multimodal explanations so non‑SQL users can verify and edit NL→SQL outputs. It uses a state‑of‑the‑art backend called rat sequence, supports editing rows and what‑if tests, and executes fixes on production. An exploratory study with 12 participants shows users detect and correct errors but sometimes see mismatches or worry about data integrity.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:15Challenges in NL2SQL systems
- 2:20DIY interactive debugging demo
- 4:31How DIY creates sample databases
- 6:53Explaining SQL execution steps
- 7:39User study findings
- 9:28Future opportunities and summary
related talks
15marchive – Manohar Sai Jasti
10mHow to test software without writing tests
15mFrom Words to Data: Using LLMs to Turn Questions into SQL
45mForget SQL! Create your own AI Data Analyst
20mExcellent Database Adventure: From Code Errors to Reliable Releases
50mLeveraging AI as a PostgreSQL DBA
10mBuilding an AI-powered SQL playground using Next.js, OpenAI, and Neon
01h 28mMaking Databases LLM-Ready Building Semantic Layers with Semantido