SchemaGraphSQL: Efficient Schema Linking with Pathfinding Graph Algorithms for Text-to-SQL on Large-Scale Databases

Name: SchemaGraphSQL: Efficient Schema Linking with Pathfinding Graph Algorithms for Text-to-SQL on Large-Scale Databases
Uploaded: 2026-10-08T03:18:06.576367+00:00
Duration: 669 s
Description: SchemaGraphSQL improves text-to-SQL by solving a critical bottleneck: linking source and destination tables in large databases. It models the schema as a graph and uses classical pathfinding algorithms to find minimal, relevant subschemas. This zero-shot method achieves state-of-the-art performance using only one lightweight LLM call per query.

SchemaGraphSQL improves text-to-SQL by solving a critical bottleneck: linking source and destination tables in large databases. It models the schema as a graph and uses classical pathfinding algorithms to find minimal, relevant subschemas. This zero-shot method achieves state-of-the-art performance using only one lightweight LLM call per query.

Open the full episode