New SAL middleware boosts Oracle NL2SQL accuracy by grounding LLMs in live schema data

Researchers have developed Schema-Aware Localisation (SAL), a new middleware for Oracle NL2SQL systems that improves SQL execution accuracy without retraining the language model. SAL integrates with Oracle databases by querying USER_TAB_COLUMNS to create a live schema map, which is then used to provide relevant table context within the LLM prompt. Additionally, SAL employs a Hallucination Index (Hidx) to validate generated SQL against the live catalog, automatically correcting predictable errors and triggering retries for more complex issues. In tests on 500 TPC-H questions using GPT-4o-mini, SAL significantly boosted execution-grounded truth from 2.2% to 62.6%, outperforming static schema hints. AI

IMPACT Enhances the reliability of LLM-generated SQL for enterprise databases, reducing errors and improving accuracy.

RANK_REASON Academic paper introducing a novel method for improving LLM performance on a specific task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• GPT-4o-mini
• Hallucination Index
• Hidx
• Oracle Autonomous Database 23c
• Oracle NL2SQL
• SAL
• Schema-Aware Localisation
• TPC-H
• USER_TAB_COLUMNS

• paper
• product
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →