ADM+S researchers named finalists in international AI retrieval challenge
Author
Date 17 June 2025
A team of researchers from the ARC Centre of Excellence for Automated Decision-Making and Society (ADM+S) at RMIT University have been selected as one of the top four finalists in an international AI competition hosted by the Technology Innovation Institute (TII) and co-located with the ACM SIGIR international conference.
The team comprising Dr Oleg Zendel, Dr Damiano Spina, Kun Ran, and Shuoqi Sun from ADM+S and colleague Dinh Anh Khoi Nguyen (RMIT)—earned first place in the Asia-Pacific session of the challenge, and are now in the top four contenders internationally.
“This opportunity allowed us to apply our research expertise to a real-world challenge, competing alongside leading research groups,” said Dr Oleg Zendel.
“We’re proud to have achieved the top result in the Asia-Pacific session as part of this effort.”
The SIGIR Live Retrieval-Augmented Generation (LiveRAG) Challenge brings together leading international research teams from both academia and industry worldwide. Organised by the TII and supported by AI71, Amazon Web Services (AWS), Pinecone, and Hugging Face, the challenge represents a major global effort to advance cutting-edge research in retrieval-augmented generation—one of the most dynamic frontiers in artificial intelligence today.
Their submission featured a system designed to maximise both ‘Relevance’, the accuracy and contextual fit of an answer, and ‘Faithfulness’, the degree to which the response is grounded in retrieved documents.
“The fact that we were able to put together our solution in such a short amount of time is a great example of the dynamic research environment we have at our information retrieval group and at ADM+S,” said Dr Damiano Spina.
“Our team, consisting of master’s and PhD students, a research fellow, and a faculty member, brings unique perspectives and expertise that contributed substantially to our success.”
Their mission was to develop AI systems capable of delivering accurate, evidence-backed responses to 500 questions within a two-hour timeframe. All participants were required to use the same AI model: Falcon 10B, an open-source large language model developed by TII.
The team was selected from a global pool of 73 applicants, with 40 teams invited to compete.
Preparation officially began in March after the team was selected, with most of the development taking place over a concentrated six-week period.
The challenge reflects growing global interest in Retrieval-Augmented Generation systems, which power AI-driven search capabilities like Google’s AI Overviews. Unlike conversational chatbots, these systems are built for single-turn question answering, combining search and natural language generation to produce high-quality, source-backed answers.
Winners will be announced live at the SIGIR 2025 LiveRAG workshop in Padua (Italy) on 17 July.
