I wrote this as a practical engineering guide to modern foundation models. It covers architectures, training systems, inference, RAG, agents, evaluation, interpretability, and safety. I would love feedback from people building or operating LLM systems.