You already know how to build software. The ladder has been renamed, not replaced.
AI engineering, as practised in real companies, is mostly excellent software engineering pointed at a strange new component: a brilliant, occasionally overconfident model that speaks JSON when asked nicely. If you're a software engineer, developer, or architect, you're not starting over - you're three-quarters of the way up a ladder that quietly got a new name. This book teaches the missing layer.
The Engine Room is a story-driven field guide with hands-on workshops. You join Falcon & Frame - a small, fictional, and entirely recognisable consultancy - as its first AI engineer, and spend twenty weeks turning a firm that runs on email attachments into one that runs on well-behaved machines. Each chapter tells the story in plain language (Pydantic is a nightclub bouncer; the API is a brilliant goldfish; prompt injection is a document that talks back), then walks you into the Workshop: real code, real commands, and exercises with solutions that wait politely until you've honestly tried.
Across 8 parts and 36 chapters you'll master: production Python foundations; how LLMs actually think (tokens, context, why hallucination is structural); choosing models and running open-weight models locally with Ollama and vLLM; prompting, structured outputs, tool use, agents, memory, and evals; MCP; retrieval-augmented generation over your own documents with pgvector and hybrid search; fine-tuning your own model with LoRA - from dataset to blind evaluation to serving; enterprise integration as a full case study; Docker, cloud deployment, CI/CD, and observability; and the privacy, security, and governance work that makes any of it trustworthy.
By the final page you'll have shipped six flagship projects and two bench projects - a portfolio of running systems, not certificates - plus a CV skills list where every line survives a follow-up question.
No machine learning background required. Sense of humour recommended.