Efficiently Linking Unstructured Data for Multi-step Reasoning

The DASE query engine enables multi-step reasoning by integrating multi-attribute filtering, vector search, and relational joins into a single execution layer. It utilizes SemJI, a sparse materialized embedding-similarity join index for rare near-neighbor pairs, to optimize thresholded similarity joins alongside structured predicate evaluation.