Preface and Introduction

Alexander Polomodov and Evgeny Sergeev begin their review of Chip Huyen's AI Engineering with practice: teams build LLM applications, use RAG, and fine-tune models, yet struggle to evaluate quality. Ten chapters lead from foundation models and evaluation to architecture, data, interfaces, and operations.

The opening chapter explains the shift from traditional ML with custom training to applications on ready-made foundation models. The hosts unpack tokens as units of representation and cost, instruction following, and multimodality. Text, images, and speech turn a model into an interface for a broad class of tasks.

Sentiment analysis, intent detection, and moderation show where an LLM replaces a narrow classifier and where it adds risk. The hype cycle separates business expectations from returns. An internal platform narrows the gap when it supplies access, observability, templates, and evaluation, not just a model.

GitHub trends and enterprise cases show growing AI tooling, while MCP and Claude Desktop point to a common protocol for context and actions. Model quality rises as cost falls. Prompt engineering simplifies prototypes, but production still demands data, verification, security, and engineering discipline.