A Multimodal RAG and ArXiv-Integrated Conversational Framework for Automated Research Discovery

Knowledge-rich fields employ multimodal documents, which require advanced analysis systems to assess document content and enable research exploration. The research introduces a unified system which combines Multimodal Retrieval-Augmented Generation (RAG) technology with automated arXiv research discovery functionality. The system operates through a user-friendly application that runs on the Streamlit web platform. The system demonstrates robust performance, achieving an average faithfulness of 0.86, with 0.81 answer relevancy on a diverse Portable Document Format (PDF) dataset. The system employs a hierarchical retrieval architecture which enhances contextual content retention capability through its complete document ingestion process that handles 15 pages within 12 seconds, which outperforms MultiModal-GPT and other evaluated systems by 33 percent. The automated arXiv Integration System also offers paper recommendations with a relevance confidence percentage of 80% at a cost of under 3 seconds. The research presents a scalable high-performance solution through its streaming conversational interface which uses LangChain Expression Language (LCEL) to demonstrate a real-world application of Conversational AI that emphasizes fast system responsiveness and user-friendly multimodal document analysis.