SAG: SQL-Retrieval Augmented Generation with Query-Time Dynamic Hyperedges

The paper introduces SAG (SQL-Retrieval Augmented Generation), a novel architecture designed to enhance retrieval-augmented generation by using SQL join queries to dynamically link events into local hyperedges at query time, thus eliminating the need for a global static graph. SAG demonstrates superior performance on multi-hop reasoning benchmarks, achieving 80.0% Recall@5 on MuSiQue and outperforms existing methods on 8 out of 9 Recall@K metrics across HotpotQA and 2WikiMultiHop. This approach allows for incremental updates and efficient scaling, making it a viable solution for practitioners needing robust retrieval mechanisms in large-scale AI applications.