This chapter explores the architecture, capabilities, and practical use of large language models (LLMs), with a particular focus on prompt engineering as a key technique for guiding model behaviour. It begins with foundational concepts, including the pre-training of language models on large corpora and the methods used to adapt them for downstream tasks. The chapter then introduces prompt engineering, detailing strategies for crafting effective prompts to elicit desired outputs and control model responses. Further sections address the challenges associated with LLMs, such as hallucinations, bias, and interpretability, and examine techniques for refining and constraining outputs. The chapter also expands the discussion beyond text generation, exploring multimodal models—particularly vision-language models—and their applications. Orchestration frameworks are introduced as tools for managing complex workflows involving multiple models and tasks. Finally, the chapter revisits core natural language processing (NLP) concepts in light of recent advances, highlighting how LLMs are reshaping the field.