RAS: Reflection-Augmented Scaling with In-Context Learning for Executable Cypher Query Generation

In three linesRAS (Reflection-Augmented Scaling) improves Cypher query generation by leveraging database error messages through in-context learning. Across three Neo4j datasets and five code-specialized language models, RAS reduces Query Execution Error Rate by 41–50% (n=5), outperforming independent resampling (32–38%).

Summary generated by Claude — human-verified