SINT-Flow: Schema Integration using Large Language Model Workflows
SINT-Flow is an LLM-based framework designed to automatically create a unified schema from multiple input tables. It uses five sequential operators—Table Splitting, Schema Matching, Table Grouping, Attribute Merging, and Integrated Schema Output—to decompose data into entity types and merge attributes. The system achieves high performance on benchmarks like SINT-Bench by combining these steps with self-consistency and reviewer loops to ensure accurate schema derivation without human intervention.