This chapter explores the transition from traditional recommendation systems to approaches powered by large language models (LLMs). It begins with an overview of conventional pipelines—covering content understanding, user modeling techniques such as collaborative filtering and matrix factorization, candidate retrieval, and ranking strategies. The chapter then highlights key limitations of traditional methods and explains how LLMs address these challenges through unified representations, enhanced personalization, and improved scalability. Various LLM-based paradigms are examined, including models that augment existing systems as well as those that serve as end-to-end recommenders. A hands-on tutorial using the MovieLens dataset illustrates this evolution in practice, comparing traditional and LLM-based methods through empirical results.