#Text-to-SQL
9 related articles
Can LLM Agents Work as Forward-Deployed Engineers? An Enterprise Text-to-SQL Reality Check
An empirical enterprise Text-to-SQL study tests whether LLM agents can serve as forward-deployed engineers: top models find over 80% of hidden requirements, but only 25% of delivery configurations pass all acceptance checks, revealing four key bottlenecks.
Schema-Free Text-to-SQL: Learning Enterprise Database Structure from Query Logs
New research proposes the first schema-documentation-free Text-to-SQL framework, building a knowledge graph from query logs and using entity-driven GraphRAG to achieve 95.7% execution validity on Beaver, far surpassing 60.8% for embedding retrieval.
Two Prompt Rules Take a Lightweight Model from 35% to 64% on Oracle Text-to-SQL
In the UniQL benchmark, lightweight Gemini 2.5 Flash with schemagate used just two Oracle dialect prompt rules to raise first-attempt accuracy from 35% to 64.5%, beating Claude Sonnet 4.5's 49.6%. This article analyzes failure attribution and evaluation insights.
Text-to-SQL Agent in Practice: Building Read-Only Safe Queries with LangChain and Gemini Free Tier
Explore a Text-to-SQL agent built with LangChain and Gemini free tier, featuring three custom tools, database-level read-only protection, and defense against prompt injection attacks.
Third-Party Benchmark Validates schemagate: RBAC for Text-to-SQL in Practice
schemagate author uses SIGMOD'27 public benchmark to test Text-to-SQL access control: 90.1 AC-F1 on BIRD beats paper's best, 5.4% wrong-answer rate, zero unauthorized reads via scope checks.
TutorialsHow Enterprise Text-to-SQL Breaks the 90% Accuracy Barrier: A Practical Guide to Agent Architecture
A deep dive into engineering strategies for enterprise Text-to-SQL to break 90% accuracy, covering precise schema retrieval, multi-Agent architecture, self-correction, and AI coding practices.
TutorialsSpring AI in Practice: Building a Text-to-SQL Agent Tool Chain from Scratch
Build a Text-to-SQL assistant using Spring AI Alibaba Agent Framework with a four-step Tool chain, plus solving ThreadLocal failures in stream mode with RunnableConfig.
Product ReviewsAix-DB Open-Source Framework: A Multi-Agent Approach to Natural Language Database Querying
Deep dive into Aix-DB: an open-source multi-agent framework using LangChain/LangGraph and MCP protocol for natural language to SQL, data analysis, and visualization.
ResearchHow Does a Semantic Layer Improve Text-to-SQL Accuracy? Latest Benchmark Reveals the Answer
A new open-source benchmark quantifies how a 4KB semantic layer boosts LLM Text-to-SQL accuracy across Claude and GPT models, validated with McNemar's test.