A Multi-Modal Retrieval-Augmented Generation System for Intelligent Document Question Answering
Year: 2026
Type: article
Abstract: This paper presents a multi-modal Retrieval-Augmented Generation (RAG) system designed to overcomelimitations of Large Language Models such as hallucination, lack of source attribution, and static knowledge.The system integrates text, images, and tabular data from documents like PDFs, Word, and Excel files. It useshybrid retrieval combining semantic embeddings and BM25 with Reciprocal Rank Fusion and cross-encoderreranking. A quality-gated caching mechanism ensures only high-quality responses are stored, reducing costand latency. Experimental results demonstrate improved accuracy (91%), reduced response time, andefficient knowledge retrieval.
Source: PRATIBODH
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7215249583