Eduardo Niglio · Eniglio Tech Press
AI Engineering Fundamentals — Eduardo Niglio
I wrote this book to help you turn AI concepts into engineering decisions. Throughout the chapters, I explore models, prompts, RAG, agents, and evaluation, connecting technical choices to quality, security, cost, and latency. I want you to learn how to experiment, analyze results, and make better-informed decisions.
About the book
I wrote AI Engineering Fundamentals to explore a question I consider essential: how can we turn the potential of artificial intelligence into systems that make sense in practice? My aim is to help you understand the decisions behind a solution, evaluate alternatives, and recognize the limits of each choice.
I start with the problem we want to solve. Before choosing a model, writing a prompt, or adding an agent, I suggest looking at the task, the available data, the expected outcome, and the conditions in which the system will be used. That clarity makes it possible to choose an approach and understand whether it is actually working.
Throughout the book, I explore models, prompt design, retrieval-augmented generation (RAG), and agents. I connect these concepts to implementation choices: what information to provide to the model, when to retrieve external content, how to integrate tools, and where to set boundaries for the system. I want you to understand why a decision is made, as well as the technique behind it.
I pay particular attention to experimentation and evaluation. A convincing answer is not necessarily a correct one. A change that improves one example can make others worse. That is why I encourage thinking in terms of hypotheses, success criteria, comparisons, and failure analysis. To me, evaluation is part of building a system: it allows us to learn from the results and choose the next step with greater confidence.
I also discuss the relationships between quality, security, cost, and latency. A solution that costs less per call may require more retries; a fast response may not meet the expected quality level. Rather than treating these dimensions in isolation, I explain why they need to be considered together, in relation to the intended use.
I see this book as a resource for people who develop software, contribute to architectural decisions, or want to better understand how AI systems are designed and evaluated. My intention is to connect the foundations with the questions that arise during implementation, without presenting any single tool or configuration as the answer to every problem.
To complement the book, I created AI Engineering Lab, a mini app with 24 practical challenges. You can analyze scenarios, choose an answer, and reflect on the reasoning behind each decision. The idea is to turn the book's concepts into an active learning experience, where both correct and incorrect answers help you identify what deserves another look.
More than a collection of techniques, I want to offer a way of thinking about AI engineering. My invitation is simple: understand the problem, experiment with purpose, evaluate carefully, and improve through evidence. That is the path I aim to explore with you throughout the book.
Eduardo Niglio
About the author
I’m Eduardo Niglio, a software engineer, AI architect, and the author of this book. I write around the questions I consider essential when designing software with AI: how to evaluate quality, protect data, and balance cost and response time. In the book, I organize these topics to help readers explain their choices. In Engineering Lab and Architecture Lab, I turn that approach into challenges and decision scenarios. My aim is to provide a foundation for thinking, experimenting, and revising solutions with care.
Reading themes
Foundations
Models, prompting, and the concepts behind AI systems.
Implementation
RAG and agents: implementation and integration decisions.
Evaluation
Evaluating responses, comparing alternatives, and interpreting results.
Governance
Security, usage limits, and responsible decisions in AI systems.
Real cases
Practical problems involving quality, cost, and latency.