International Journal of Scientific Research and Engineering Development

The growing amount of unstructured information contained in the heterogeneous forms of storage, including documents, pictures and audio recordings presents huge problems to the conventional retrieval systems that rely on keywords. Retrieval-Augmented Generation (RAG) enhances response grounding by combining retrieval processes with Large Language Models (LLMs); most of the existing systems are textbased and require cloud access. The article describes the design and implementation of a multimodal RAG framework system based offline to perform unified semantic search. The suggested system consumes PDF, DOCX, image, as well as audio files and transforms them into a common embedding space, and the similarity-based retrieval across the modalities. An offline LLM makes context sensitive responses that are purely based on the retrieved information, and their responses are provided with references to the sources to make it transparent. Experimental analysis proves the cross-modal retrieval, the lessening of hallucination and safe functioning without internet addiction. The architecture is especially adapted to privacy sensitive enterprise and government environments where explainable offline intelligence is needed.

Sandeep Kulkarni, Aiman Patel, Sandesh Singh,"Design and Implementation of an Offline Multimodal Retrieval Augmented Generation System for Unified Semantic Search" International Journal of Scientific Research and Engineering Development, V9(2): Page(1721-1729) Mar-Apr 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.