BDM based smart library browsing and dialoggue system

This study comprehensively examines the technical infrastructure, architectural design, and performance evaluation of an AI-supported intelligent library search and dialogue system, specifically developed for Batman University based on a Large Language Model (LLM). Considering the limitations of traditional OPAC (Online Public Access Catalog) systems and the growing expectations of users, contemporary information retrieval techniques in the literature have been thoroughly analyzed. In particular, vector-based data structures, Retrieval-Augmented Generation (RAG) architecture, model enhancements through Reinforcement Learning with Human Feedback (RLHF), ensemble learning strategies, historical inconsistency detection, and Explainable AI (XAI) methodologies are elaborated in detail. The system is implemented using the Python programming language and the LangChain framework to facilitate multi-source information retrieval. The primary data sources include Batman University’s YORDAM catalog, followed by its institutional repository (e-Archive), and lastly, a web search module supported by ChatGPT. This multi-stage query pipeline is designed to optimize the information retrieval process and provide access to comprehensive, accurate, and up-to-date data. The Chroma-based vector database further improves system performance by enabling efficient memory management and semantic search capabilities. Moreover, user feedback is continuously collected and utilized within the RLHF framework to support adaptive learning and ongoing model refinement. Consequently, the system demonstrates a dynamic and evolving structure in terms of both technical functionality and user interaction. Advanced analytical components, such as historical inconsistency detection, contribute to information reliability, while Explainable AI techniques aim to enhance transparency and interpretability of system outputs for end-users and library personnel. Findings from this research indicate that the developed system significantly improves information retrieval speed, enhances the precision of search results, and increases overall user satisfaction. As an integrated and sustainable AI-based library solution, the system represents a promising model for academic libraries in Turkey, offering valuable contributions to the efficiency and effectiveness of scholarly research processes.