Two RAG wins in one year: LiveRAG and MMU-RAG

Our team had two major RAG competition results in 2025.

First, we won the SIGIR 2025 LiveRAG Challenge. Later, at NeurIPS 2025 MMU-RAG, our system won the Best Dynamic Evaluation award in the Open Source category for the text-to-text track, as described in our MMU-RAG system paper.

These results came from two related systems developed in the same year: GRAG for LiveRAG, and R2RAG for MMU-RAG.

The two results

1) SIGIR 2025 LiveRAG Challenge

On the official LiveRAG challenge page, the winners are listed as:

First place: RMIT-ADMS — Kun Ran, Shuoqi Sun, Khoi Nguyen Dinh Anh, Damiano Spina, Oleg Zendel

The same page states that on the live challenge day, teams had to answer a stream of unseen questions under a two-hour time limit, and 25 teams returned valid answers.

The ADM+S announcement reports that the competition drew 70 teams from 27 countries, and that during the live event teams had to answer 500 never-before-seen questions using the same AI model and dataset.

2) NeurIPS 2025 MMU-RAG

A later system, R2RAG (Routing-to-RAG), is described in our paper RMIT-ADM+S at the MMU-RAG NeurIPS 2025 Competition. In that paper, the system is described as an award-winning approach for the Text-to-Text track, and the paper states that it won the Best Dynamic Evaluation award in the Open Source category.

The ADM+S announcement on MMU-RAG describes this as the team’s second major RAG competition result that year. It also reports that 81 teams registered, but only 8 submitted a fully working system.

What we built

The second system built on ideas developed in the first.

LiveRAG: G-RAG / GRAG

In our LiveRAG paper, we described Generation-Retrieval-Augmented Generation (GRAG).

The method works in three main steps:

• generate a hypothetical answer first,
• use that generated answer alongside the original question during retrieval,
• then apply LLM-based pointwise re-ranking before final answer generation.

As Damiano’s publication page summarizes, the submitted system achieved the highest Borda score, based on aggregated manual evaluation of Coverage, Relatedness, and Quality, and ranked first in the SIGIR 2025 LiveRAG Challenge.

MMU-RAG: R2RAG

In MMU-RAG, we extended these ideas with Routing-to-RAG (R2RAG).

In the MMU-RAG paper, R2RAG is described as a research-focused RAG architecture composed of lightweight components that dynamically adapt retrieval strategy based on:

• inferred query complexity, and
• evidence sufficiency.

The paper also states that the system uses smaller LLMs and can operate on a single consumer-grade GPU, while still supporting complex research tasks.

Why these results matter technically

Taken together, the two systems highlight several recurring design themes:

• retrieval quality matters;
• evidence use matters;
• evaluation design matters;
• latency and robustness matter;
• and careful system design can beat brute force.

They also show a progression from a competition-winning LiveRAG pipeline to a more adaptive RAG architecture designed for dynamic evaluation settings.

Team

These results were team efforts.

For LiveRAG, the official challenge page lists the winning team as:

• Kun Ran
• Shuoqi Sun
• Khoi Nguyen Dinh Anh
• Damiano Spina
• Oleg Zendel

For MMU-RAG, the paper lists the authors as:

• Kun Ran
• Marwah Alaofi
• Danula Hettiachchi
• Chenglong Ma
• Khoi Nguyen Dinh Anh
• Khoi Vo Nguyen
• Sachin Pathiyan Cherumanal
• Lida Rashidi
• Falk Scholer
• Damiano Spina
• Shuoqi Sun
• Oleg Zendel