Structured Text-to-SQL pipeline combining LLM sketch generation, schema-aware two-phase retrieval, and validation-driven refinement. Improves SQL controllability and correctness via schema embeddings, join-path constraints, and minimal-edit correction. Research-oriented, fully traceable POC on MiniDev / Spider-style datasets.
nlp information-retrieval research sql-generation structured-search text-to-sql schema-aware llm prompt-engineering retrieval-augmented-generation llm2sql two-phase-retrieval
-
Updated
Jan 12, 2026 - Python