A SYSTEMATIC REVIEW OF VECTOR DATABASE USE IN RETRIEVAL-AUGMENTED GENERATION FOR LLM-BASED EDUCATIONAL PLATFORMS

Abstract

This systematic review explores the use of vector databases in Retrieval-Augmented Generation (RAG) for educational platforms based on large language models (LLMs). As RAG becomes a promising approach to enhance the contextual accuracy of LLM outputs by retrieving relevant content, vector databases serve as a core component for storing and retrieving embedded educational materials. This review is comprised of 9 studies from 2023 to 2025, focusing on use cases in higher education, including domain specific applications and chatbots for student and educator support. Findings show diverse choices of vector stores, such as FAISS, Chroma, Qdrant, Weaviate, Milvus, Vectara, MongoDB and Postgres with pgVector, often combined with orchestration frameworks like LangChain or LlamaIndex. The reporting on embedding models, orchestration frameworks and system architecture is inconsistent, limiting the comparability of studies and reducing confidence in synthesizing performance trends, which impacts the reliability of conclusions drawn from the review. The findings provide a reference point for researchers and developers creating context-aware, LLM-based educational platforms, and suggest future research directions including performance benchmarking, model transparency, and evaluating learning outcomes.