Retrieval Augmented Generation, or RAG, offers a practical solution for large language models to access specific, up-to-date information. It bypasses the need for expensive and time-consuming model retraining by incorporating relevant data directly into the AI's response process. The core of RAG involves first retrieving pertinent pieces of your own documents before the AI generates an answer. This process consists of three main components: chunking, embeddings, and vector search. Chunking breaks down large documents into smaller, manageable segments to facilitate efficient processing. Embeddings convert these text chunks into numerical representations, allowing computers to understand semantic similarity. Computers cannot directly compare meaning, so each chunk is turned into a list of numbers called an embedding. Text with similar meaning results in mathematically close embeddings, even with different wording. A SHA-256 hash-based embedding cache was implemented to reuse existing embeddings, significantly reducing processing time and cost. Vector search, using libraries like FAISS, quickly finds the most relevant chunks based on their numerical embeddings compared to the user's query. The entire RAG flow involves ingesting documents, chunking them, creating embeddings (with caching), storing these in a vector index, embedding user questions, retrieving matching chunks, and then feeding both the question and chunks to the AI for an informed answer. This approach ensures answers are grounded in original sources, scalable, and easily updated without model retraining. Effective chunk size and robust retrieval are crucial for successful RAG implementation.