Abstract: 

Recent advancements in information retrieval have revolutionized the way systems retrieve, rank, and generate knowledge-based responses. This talk presents three recent research contributions that address key challenges in retrieval-augmented generation (RAG) and re-ranking techniques. First, we introduce ASRank, a zero-shot re-ranking approach leveraging answer scent—a novel method that enhances ranking by aligning retrieved passages with expected answer patterns, significantly improving retrieval accuracy. Next, we explore DynRank, a dynamic zero-shot prompting framework that classifies questions into fine-grained types to generate contextually relevant prompts, optimizing passage retrieval for open-domain question answering. Finally, we present Rankify (https://github.com/DataScienceUIBK/Rankify), a comprehensive Python toolkit that unifies retrieval, re-ranking, and RAG into a modular framework, enabling researchers to experiment with diverse ranking models in a standardized environment. Through an in-depth analysis of these approaches, we highlight the evolving landscape of information retrieval and its implications for modern search and AI-driven question-answering systems.