Retrieval-Augmented Generation with Large Language Models

Large Language Models (LLMs) have shown impressive capabilities in generating human-like responses, but they often rely solely on pre-trained knowledge, leading to outdated or inaccurate outputs. Retrieval-Augmented Generation (RAG) addresses this limitation by combining LLMs with external knowledge retrieval, enabling models to access and use up-to-date, domain-specific information during response generation. This thesis explores the architecture and components of a RAG system, including retrievers, vector databases, embedding models, and language model integration. The goal is to understand how RAG enhances the accuracy and reliability of generative systems in real-world scenarios. The work includes a conceptual and technical analysis of the RAG pipeline, supported by a small-scale implementation using open-source tools and custom data. Potential applications in areas such as document-based question answering, internal knowledge access, and AI assistants are discussed, along with the benefits and limitations of this approach.