A Semantic Retrieval and Generation Framework for Alumni Intelligence Using LLaMA 3

Abstract

The study focused on the development of the Alumni Intelligence System, a scalable and intelligent solution designed to preserve and enable access to alumni project data within an academic institution. The system addresses significant challenges for limited and outdated historical records, with 155 usable records by integrating LLaMA 3, a large language model, with Retrieval- Augmented Generation (RAG). A vector database is utilized to hold the vector embeddings created from structured alumni project data. Upon receiving a query, the system performs similarity search to retrieve relevant contexts, which are then used to generate accurate and context-aware responses via LLaMA 3. Performance evaluation using metrics such as cosine similarity (0.78), precision (0.79), and BLEU score (0.64) indicates the system’s effectiveness in retrieving and generating relevant academic information. The proposed system demonstrates the potential of LLM-powered architectures in enhancing academic knowledge retention, research support, and alumni data utilization.