GraphSense-RAG: An Intelligent Multi-Modal Document Question Answering System

Abstract: The rapid growth in the digital documents made a critical challenge in retrieving accurate information from the large volumes of data. This question-answering system has become a challenging task for the companies and individuals who deals with the huge volume of information. To overcome these challenges, the proposed mainly focuses on Document Question Answering system based on Retrieval-Augmented Generation approach which is integrated with the embedding search. The main aim of this project is to provide accurate and reliable information for the user query from the document. The proposed system supports multi modal data processing and also uses hybrid retrieval technique which includes semantic search, keyword matching and metadata filtering. These techniques are used to enhance the retrieval performance and accuracy. The proposed system also included with a dedicated hallucination detection mechanism which validates the generated result by grounding them in retrieved evidence. In addition to this the system also supports automatic document summarization and indexing. It also allows the user for efficient ingestion of new documents. The proposed system provides scalable, reliable solution which can be effectively used for academic research and also for the enterprise companies.

Index Terms: Retrieval-Augmented Generation, Vector Embedding Search, Multi-Modal Question Answering, Hybrid In- formation Retrieval, Knowledge Graph Integration