MASS-RAG: Multi-Agent Synthesis Retrieval-Augmented Generation

TL;DR

MASS-RAG introduces a multi-agent framework for retrieval-augmented generation, enhancing evidence processing and synthesis to improve accuracy when handling noisy or heterogeneous retrieved data.

Contribution

It proposes a novel multi-agent synthesis architecture that structures evidence handling into specialized roles, improving over traditional single-process RAG models.

Findings

Consistently outperforms strong RAG baselines on four benchmarks.

Improves performance especially when relevant evidence is distributed across contexts.

Exposes multiple evidence views for better comparison and integration.

Abstract

Large language models (LLMs) are widely used in retrieval-augmented generation (RAG) to incorporate external knowledge at inference time. However, when retrieved contexts are noisy, incomplete, or heterogeneous, a single generation process often struggles to reconcile evidence effectively. We propose \textbf{MASS-RAG}, a multi-agent synthesis approach to retrieval-augmented generation that structures evidence processing into multiple role-specialized agents. MASS-RAG applies distinct agents for evidence summarization, evidence extraction, and reasoning over retrieved documents, and combines their outputs through a dedicated synthesis stage to produce the final answer. This design exposes multiple intermediate evidence views, allowing the model to compare and integrate complementary information before answer generation. Experiments on four benchmarks show that MASS-RAG consistently improves performance over strong RAG baselines, particularly in settings where relevant evidence is distributed across retrieved contexts.