Volume 2 - Issue 3, May - June 2026

Recent advancements in Artificial Intelligence and Large Language Models (LLMs) have significantly improved conversational systems and intelligent information retrieval applications. However, most modern AI assistants rely on cloud-based infrastructures, leading to concerns regarding data privacy, API dependency, operational cost, and internet connectivity. This paper presents the development of an Autonomous Knowledge Assistant using Hybrid Vector Search integrated with Retrieval-Augmented Generation (RAG) for secure and offline conversational information retrieval. The proposed system combines dense semantic retrieval using FAISS with sparse keyword-based retrieval using BM25 to improve response accuracy and contextual relevance. The architecture integrates Streamlit for the user interface, LangChain for retrieval orchestration, Ollama for local LLM deployment, and FAISS for vector indexing. The system also includes query rewriting, exact answer extraction, response polishing, and administrative knowledge management functionalities. The proposed assistant enables users to interact with institutional knowledge bases using natural language queries without requiring technical expertise. Experimental evaluation demonstrates that hybrid retrieval significantly improves precision and reduces hallucinated responses compared to standalone retrieval methods. The developed system provides a cost-effective, scalable, privacypreserving, and fully offline conversational AI solution suitable for educational institutions and enterprise knowledge systems.

Peese Vamshi, Dr.D.William Albert, G Sree Ramulu,"An Autonomous Knowledge Assistant Using Hybrid Vector Search and Ollama" International Journal of Advanced Multidisciplinary Research and Educational Development, V2(3): Page(614-617) May-June 2026. ISSN: 3107-6513. www.ijamred.com. Published by Scientific and Academic Research Publishing.