FALCON Framework Generates Ambiguity-Aware Synthetic NL2SQL Training Data

FALCON uses SQL seeding and persona-based prompting to generate complex, ambiguity-aware NL-to-SQL pairs using compact open models. An alignment-based filtering mechanism ensures the generated data maintains structural complexity and high quality for training purposes.