Modern Search for AI Professionals

Modern Search for AI Professionals is the working title of a technical book I am writing about how to build, evaluate, and improve search systems. The book is at an early stage. I am settling its scope and chapter outline, and no chapters have been drafted yet.

The problem

Many AI engineers now build retrieval into their products through RAG pipelines and vector databases, often without a background in information retrieval. When results are poor, it can be hard to tell whether the cause lies in the query, the catalog, the index, the ranking, or the way quality is being measured. This book is meant to close that gap.

Who it is for

The book is written for AI professionals who build or lead search-enabled products. I expect a mix of machine learning engineers, AI engineers who have built RAG or vector search applications and want to understand the search system beneath them, and technically skilled product leaders. It assumes familiarity with APIs, embeddings, and basic machine learning, and builds lexical retrieval, ranking, evaluation, and production search mechanics from a practitioner's approach.

The premise

Modern search is an iterative, measurable system. A team should start from a sound lexical baseline, then add semantic retrieval, learned ranking, calibrated decisions, and operational complexity only when evidence shows that each layer earns its place. Product search is the running example, because it combines short and ambiguous queries, structured catalogs, and tight latency budgets.

The candidate topics include relevance evaluation and judgment collection, query and catalog understanding, lexical, learned sparse, dense, hybrid, and multi-vector retrieval, adapting retrieval models with training data, ranking and result-set composition, serving trade-offs, and feedback loops for continuous improvement. Agentic search for complex product research is planned as a major advanced topic.

Where it stands

So far I have defined the audience, the thesis, and the anchor domain. The scope draws primarily on my own experience as a search and machine learning practitioner, complemented by a review of the topic landscape that drew on recent information retrieval research. The next step is choosing among alternative chapter outlines. Companion code with reproducible experiments is planned but has not started.

The book is a sibling to AI-Powered Personalization, my book on recommender systems. Each is written to stand on its own.