AI Engineering

Building an application on top of a foundation model no longer takes a background in machine learning. Models are now offered as a service, so developers with little or no AI experience can put them to work. This guide from Chip Huyen treats AI engineering as its own discipline: the practice of making applications from models that already exist.

It opens with an overview of the field, setting it against traditional ML engineering and laying out the new AI stack. Evaluation gets particular weight, because wider use of AI means more chances of serious failure. The book covers ways to assess open-ended models, including the fast-growing practice of using one AI model to judge another.

The core is a framework for developing an application, starting with simple techniques and moving toward more sophisticated ones. Model adaptation methods are covered in turn: prompt engineering, RAG, fine-tuning, agents and dataset engineering, with attention to how and why each works. Readers also learn how to choose models, datasets, benchmarks and metrics, and how to find and reduce latency and cost bottlenecks when serving foundation models.

This is a practical technical handbook for application developers. Huyen previously wrote Designing Machine Learning Systems (O'Reilly), and this book is complementary to it. She has worked at Voltron Data, Snorkel AI and NVIDIA, and has taught at Stanford.