LLM Engineer's Handbook vs AI Engineering: which should you read?
Short answer
Read AI Engineering if you want the principles of building on foundation models, such as evaluation, prompting, RAG, agents and fine-tuning, explained without tying you to tools. Read LLM Engineer's Handbook if you learn by building: it takes one LLM system from data collection through fine-tuning, RAG, deployment on AWS and monitoring, in code. They complement each other, and many engineers read AI Engineering first.
By Dr. Dulan Dias, co-founder and CTO of Tellme AI
Key facts
- LLM Engineer's Handbook, by Paul Iusztin and Maxime Labonne, was published by Packt in October 2024.
- LLM Engineer's Handbook builds one project, an LLM Twin, end to end, using ZenML for pipelines and AWS SageMaker for training and serving.
- AI Engineering, by Chip Huyen, was published by O'Reilly in December 2024 and is organised around adapting foundation models rather than a codebase.
- Hands-On Large Language Models, by Jay Alammar and Maarten Grootendorst, was published by O'Reilly in September 2024 and teaches through hundreds of custom figures and runnable notebooks.
- A 2026 reading list for developers on travis.media orders them as Hands-On Large Language Models, then AI Engineering, then LLM Engineer's Handbook.
- LLM Engineer's Handbook's repository estimates that running its project once with the default settings costs roughly 25 US dollars, mostly on AWS SageMaker.
Definitions
- LLM Twin
- The project LLM Engineer's Handbook builds: a language model fine-tuned and grounded on one person's writing, used as the thread that connects every chapter's pipeline.
- LLMOps
- The operating practices for language model applications: versioning prompts, data and models, automated pipelines, evaluation, deployment, monitoring and cost control.
- Feature, training and inference pipelines
- A way of splitting an ML system into three independent parts: one that prepares data, one that trains models, and one that serves predictions. LLM Engineer's Handbook is built around this split.
- Preference alignment
- Further training a model on pairs of better and worse responses so that it favours the kind of output people prefer. Direct preference optimisation (DPO) is a common method.
How do the books compare?
| LLM Engineer's Handbook | AI Engineering | Hands-On Large Language Models | |
|---|---|---|---|
| Authors | Paul Iusztin, Maxime Labonne | Chip Huyen | Jay Alammar, Maarten Grootendorst |
| Publisher, date | Packt, October 2024 | O'Reilly, December 2024 | O'Reilly, September 2024 |
| Approach | One system built end to end in code | Principles and frameworks, little code | Concepts through figures and notebooks |
| Stack | Python, ZenML, AWS SageMaker and more | Tool-agnostic | Python notebooks, open models |
| Strongest on | Pipelines, fine-tuning, RAG, deployment | Evaluation, prompting, RAG, agents, architecture | How LLMs work and how to use them |
What LLM Engineer's Handbook is like
The Handbook teaches by construction. One project runs through the book: collecting data, building a RAG feature pipeline, supervised fine-tuning and preference alignment, evaluating the model, optimising inference, deploying the inference pipeline, and wiring it all together with MLOps and LLMOps practices. Its value is seeing every stage connected, which most books only describe.
Two things to expect. The stack is specific, so some tool choices will date faster than the structure; and the full project runs on a paid cloud account, so budget for it or run the parts you need locally. Readers who skip the cloud steps still get the architecture.
What AI Engineering is like
AI Engineering is the map rather than the route. It explains how foundation models behave, how to evaluate open-ended output, when to use prompting, RAG, agents or fine-tuning, how to engineer datasets, and how to make inference faster and cheaper. Each topic is framed as a decision with trade-offs, which is why engineers return to it when the next model arrives.
Its evaluation chapters are the strongest reason to read it: they explain how to judge open-ended output, including when using a model as a judge is reasonable and what its limits are.
It is weaker exactly where the Handbook is strong: it will not walk you through building and deploying a system. The comparison with Designing Machine Learning Systems covers how it relates to Huyen's earlier book.
Where Hands-On Large Language Models fits
Hands-On Large Language Models is the gentlest of the three. It explains tokens, embeddings and the inside of a transformer visually, then applies pretrained models to classification, clustering, semantic search, RAG and fine-tuning in notebooks you can run. It suits engineers who want intuition before architecture, and it is often read first.
Which order should you read them in?
A practical rule: read one book for judgement and one for practice. AI Engineering supplies the judgement; either of the other two supplies the practice, depending on whether you want to understand models (Hands-On Large Language Models) or ship a pipeline (LLM Engineer's Handbook). Reading all three back to back before building anything is the slowest way to learn from them.
| You want to | Read | Then |
|---|---|---|
| Understand LLMs before building | Hands-On Large Language Models | AI Engineering |
| Make sound product decisions about LLM features | AI Engineering | LLM Engineer's Handbook |
| Build a full pipeline to learn the moving parts | LLM Engineer's Handbook | AI Engineering |
| Fine-tune open models | LLM Engineer's Handbook | Hands-On Large Language Models |
| Ship an LLM feature inside an existing product | AI Engineering | From Model to Product (disclosed below) |
What none of the three is about
All three books treat the language model as the centre of the system. Many products use one as a component instead: rules and a trained classifier make the decision, and the language model explains it, drafts text or answers questions, with a fallback when it is slow, expensive or wrong. Deciding where a language model belongs at all is the step before any of these books begin.
From Model to Product starts there. Its chapter on choosing an approach ranks rules, search, classical ML, deep learning and language models by cost and risk, and its running product uses a language model only to explain decisions that rules and a vision model make, served with FastAPI and containers on any cloud. It does not cover fine-tuning or agents in depth. Disclosure: From Model to Product is written by Dr. Dulan Dias, who also writes this site. It appears where it fits, with what it leaves out, and is not ranked above the other books.
Frequently asked questions
Is LLM Engineer's Handbook worth reading?
Yes, if you learn by building. It is one of the few books that connects data collection, fine-tuning, RAG, deployment and monitoring in a single working codebase. Expect a specific stack, including AWS, and treat its tool choices as examples. Readers who want principles rather than a build may prefer AI Engineering.
Do you need AWS to follow LLM Engineer's Handbook?
The book's full project deploys to AWS, using SageMaker, and its repository estimates roughly 25 US dollars for one run with the default settings. You can follow the architecture, the data pipelines and much of the fine-tuning and RAG work without deploying, and adapt the serving steps to another platform.
Which book is better for learning RAG?
For deciding when and how to use retrieval, AI Engineering, which covers RAG alongside agents, prompting and fine-tuning as alternatives. For building a retrieval pipeline in code, LLM Engineer's Handbook, which implements one as part of its project. Hands-On Large Language Models explains semantic search and RAG with runnable notebooks.
Which book covers fine-tuning in more depth?
LLM Engineer's Handbook covers fine-tuning hands-on, including supervised fine-tuning and preference alignment on its project. AI Engineering explains when fine-tuning is worth it and how parameter-efficient methods work, and Hands-On Large Language Models covers fine-tuning in notebooks.
Is LLM Engineer's Handbook good for beginners?
It suits engineers comfortable with Python and the basics of machine learning, because it moves quickly through a real stack. Beginners usually find Hands-On Large Language Models an easier first book, then come to the Handbook once they want to build a complete system.
Should I read Hands-On Large Language Models before the other two?
If tokens, embeddings and attention are still abstract to you, yes. It builds the intuition the other two assume, and its notebooks let you test each idea. If you already work with LLM APIs daily, you can start with AI Engineering instead.
Which of these books is the most up to date?
All three were published within four months of each other in late 2024, so none covers models released since. AI Engineering ages best because it is organised around decisions rather than tools. The Handbook's specific stack dates fastest, though its pipeline structure does not.
What does From Model to Product add to these books?
From Model to Product, by this site's author Dr. Dulan Dias, covers the system around a language model: deciding whether one is needed, combining it with rules and trained models, serving it with FastAPI and containers, and monitoring cost and reliability. It does not replace these books on fine-tuning, evaluation or agents.
Sources
About the author
Dr. Dulan Dias is co-founder and Chief Technology Officer of Tellme AI in Belfast; founder of EchonLabs, grown from 2017 to more than 60 people before its acquisition in 2026; holder of a PhD in AI and data science from Queen’s University Belfast; and the author of From Model to Product.