schema-linking

[EMNLP 2025 Main] LinkAlign: Scalable Schema Linking for Real-World Large-Scale Multi-Database Text-to-SQL

Squrve is a lightweight yet powerful framework for translating natural language into SQL over complex databases.

Text-to-SQL + RAG based intelligent data assistant with schema linking and self-reflection.

Adaptive prompting for Text-to-SQL: selective schema and value augmentation, column retrieval, and relation discovery.

Draft SQL and multi-candidate Text-to-SQL pipeline with schema compression, repair, and selector

Automatic metadata extraction for text-to-SQL — schema linking pipeline reproducing the AT&T method, evaluated on BIRD.

A database's schema grounded in its data and described in sentences, ready for a text to SQL agent to read

Schema linking for text-to-SQL as graph search: catalog metadata (DDL, dbt, Unity Catalog, Glue, Collibra) in, join-complete sub-schema out as annotated DDL. HTTP API, MCP server, web UI. No embeddings, no LLM required.

LoRA fine-tuned Qwen2.5-Coder-1.5B for schema linking in NL-to-SQL: given a question and a database schema, predicts the referenced tables and columns as JSON. Uses self-consistency decoding (k=10) and post-processing to drop hallucinated identifiers.

Academic Text-to-SQL project using RAG, FAISS, FastAPI, schema linking, SQL repair, and semantic validation.

Text-to-SQL system achieving 93.7% execution accuracy on Spider benchmark using fully local LLMs. 11-step adaptive pipeline with three-layer schema linking, complexity routing, DAIL-SQL reranking, and validation-feedback retry. Includes SFT + GRPO fine-tuning. CSE 573 Semantic Web Mining project.

LoRA fine-tuned Qwen2.5 for text-to-SQL schema linking: lexical schema retrieval + ID-constrained decoding

Engineering-focused AI data analysis agent with schema linking, safe Text-to-SQL, self-correction, semantic critic, SSE streaming, and ECharts.

Text2SQL agent system for the JYDB financial database: ReAct-style agent with schema linking, read-only SQL tools, and strict Binary EX evaluation.

Self-correcting natural-language-to-SQL agent for SQLite: schema linking, read-only safety validation, and an execution-feedback repair loop, with an execution-accuracy eval.

零依赖、可离线运行的 Text2SQL 全链路演示：Schema Linking 三阶段 + SQL 安全校验 + 执行级评估（Execution Accuracy）

Ask a database questions in plain English and get SQL that actually runs. The key is schema linking: relevant tables are retrieved by embedding similarity (with foreign-key expansion) before prompting, and quality is measured by execution accuracy (do the returned rows match?) not by string-matching SQL.

Natural-language analytics agent for arbitrary PostgreSQL databases: hybrid NL→SQL (learned / deterministic / coder cascade, no router), document RAG with citations, whitelist-restricted web research. Zero code for a new schema.

Official implementation of LLM-guided semantic and structural priors for Text-to-SQL.

Database Entity Recognition: grounding natural-language questions to database tables, columns, and values. Code, trained-model artifacts, and results for the DB-ER benchmark (IEEE IRI 2025); dataset on the Hugging Face Hub.