The team RAGtifier from L3S secured third place in the SIGIR 2025 LiveRAG Challenge. The Technology Innovation Institute (TII) hosted the competition with support from AWS, Pinecone, AI71 and Hugging Face. It focused on question answering (QA) through retrieval-augmented generation (RAG).
Participants were tasked with building an RAG system using a fixed corpus (FineWeb-10BT) and the open-source Falcon3-10B-Instruct LLM. On the crucial LiveChallenge Day, held on 12 May, a total of 70 invited teams from academia and industry across 27 countries provided solutions to 500 unseen questions within a two-hour time frame.
RAGtifier (William Xion, Hailay Teklehaymanot, Oleh Astappiev and Tim Cofala) focused on evaluating and optimising state-of-the-art RAG solutions that target retrieval and reranking, as well as iterative generation approaches, delivering high performance in both the automated and manual evaluation stages. The L3S team secured a place among the finalists with this approach, ultimately coming in third behind Magikarp from China and RMIT-ADMS from Australia.
The award ceremony took place during the LiveRAG workshop on 17 July 2025 in Padua (Italy) as part of the SIGIR Congress.
RAGtifier Report and Code:
https://arxiv.org/abs/2506.14412
https://git.l3s.uni-hannover.de/liverag/ragtifier
LiveRAG Challenge Report:
https://arxiv.org/abs/2507.04942
