MAGDA: A Clustering-Based Retrieval-Augmented Generation System for Gender Gap Analysis in Digital Libraries

This work introduces MAGDA (Multi-document Aggregation via Global Document-level clustering Architecture), a domain-specific RAG system that uses a clustering-based chunking and retrieval strategy to capture semantic relations across documents in the Gender*More corpus. By aggregating inter-document information, MAGDA improves the grounding and relevance of generated answers. We evaluate the system on a curated set of real queries and validated responses, showing that domain-adapted RAG pipelines combined with cross-document processing significantly enhance information access and explainability in specialized scientific digital libraries.