Rama Ramakrishnan teaches session nine of MIT's 15.773 Hands-On Deep Learning, continuing from the transformer encoder architecture covered in the previous class. He walks through how large language models predict the next word in a sequence, building up the mechanics that let transformers generate coherent text. From there the lecture turns to Retrieval Augmented Generation, explaining how retrieval systems feed external documents into a language model's context so it can answer questions grounded in specific source material rather than relying only on what it memorized during training. The session is part of MIT's Sloan School curriculum on applied deep learning for management and business students, and treats LLMs and RAG as practical tools rather than abstract theory, with an eye toward how these systems get built and deployed in real applications.