Evolutionary Paradigms in University Knowledge Management: A Comparative Analysis of Intent Based Architectures and Vectorized Semantic Retrieval

Abstract

This research examines the transition of conversational artificial intelligence within higher education administrative systems, specifically focusing on the shift from traditional intent-based architectures to vectorized semantic retrieval models. By analyzing the current implementation at Singidunum University, which utilizes the Rasa framework and a Dual Intent and Entity Transformer (DIET) pipeline, the study identifies a "knowledge bottleneck" when dealing with unstructured regulatory corpora. The investigation compares this deterministic model with emerging vector database technologies such as Pinecone and Weaviate, which employ high-dimensional embeddings and Cosine Similarity to retrieve information based on semantic proximity. The results suggest that while intent-based systems provide high precision for frequent queries, a hybrid model incorporating Retrieval-Augmented Generation (RAG) is essential for navigating complex institutional statutes.