RMIT-ADM+S at the MMU-RAG NeurIPS 2025 Competition

2026-02-24

AI Summary

This work proposes Routing-to-RAG (R2RAG), a novel architecture designed to efficiently support complex research tasks in resource-constrained settings through retrieval-augmented generation (RAG). The approach introduces a dynamic routing mechanism that adaptively selects between retrieval and generation pathways based on query complexity and evidential sufficiency, leveraging lightweight components and small-scale large language models. A qualitative output analysis module further refines generation quality, enabling high-performance inference on a single consumer-grade GPU. The system was awarded the Best Dynamic Evaluation Prize in the Open-Source category of the NeurIPS 2025 MMU-RAG Challenge Text-to-Text Track, demonstrating its effectiveness and efficiency.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

This paper presents the award-winning RMIT-ADM+S system for the Text-to-Text track of the NeurIPS~2025 MMU-RAG Competition. We introduce Routing-to-RAG (R2RAG), a research-focused retrieval-augmented generation (RAG) architecture composed of lightweight components that dynamically adapt the retrieval strategy based on inferred query complexity and evidence sufficiency. The system uses smaller LLMs, enabling operation on a single consumer-grade GPU while supporting complex research tasks. It builds on the G-RAG system, winner of the ACM~SIGIR~2025 LiveRAG Challenge, and extends it with modules informed by qualitative review of outputs. R2RAG won the Best Dynamic Evaluation award in the Open Source category, demonstrating high effectiveness with careful design and efficient use of resources.

Problem

• retrieval-augmented generation
• query complexity
• evidence sufficiency
• resource-constrained inference
• dynamic retrieval

Innovation

• Routing-to-RAG
• retrieval-augmented generation
• dynamic retrieval strategy
• lightweight LLMs
• query complexity inference