What are the best books on deploying machine learning models?
Short answer
The best books on deploying machine learning models in 2026 are Designing Machine Learning Systems for the decisions behind a deployment, Christian Kästner's Machine Learning in Production for the engineering discipline, Machine Learning Production Systems for pipelines, Reliable Machine Learning for operating models, and LLMs in Production or LLMOps for language models. Several popular MLOps books predate LLM applications, so check the year before you buy.
By Dr. Dulan Dias, co-founder and CTO of Tellme AI
Key facts
- Christian Kästner's Machine Learning in Production: From Models to Products was published by MIT Press in April 2025, and its full text is online under a Creative Commons licence.
- Machine Learning Production Systems, by Robert Crowe, Hannes Hapke, Emily Caveness and Di Zhu, was published by O'Reilly in 2024.
- Reliable Machine Learning (O'Reilly, 2022), by Cathy Chen, Niall Richard Murphy, Kranti Parisa, D. Sculley and Todd Underwood, applies site reliability engineering principles to machine learning.
- LLMs in Production, by Christopher Brousseau and Matthew Sharp, was published by Manning in December 2024.
- Andriy Burkov's Machine Learning Engineering (2020) is distributed on a read first, buy later principle.
- Introducing MLOps (2020), Practical MLOps (2021) and Machine Learning Design Patterns (2020) were all published before language model applications became mainstream.
Definitions
- Model deployment
- Making a trained model available to the software that needs its predictions, as an online service, a batch job or a model running on a device, together with the release process around it.
- MLOps
- The practices that join building ML systems to operating them: automated pipelines, versioning of code, data and models, testing, deployment, monitoring and retraining.
- Online and batch inference
- Online inference answers each request as it arrives, usually behind an API. Batch inference scores many records on a schedule and stores the results for later use.
How these books were chosen
Deploying a model is less about one command than a run of decisions: how predictions are served, how releases are tested and rolled back, how the model is watched once real data arrives, and who is woken when it drifts. The books below each own part of that run. Each entry gives the publisher and year, and flags whether it was written before language model applications took off in 2023, because that changes what it covers, not whether it is good.
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.
The list at a glance
| Book | Author(s) | Publisher, year | Before LLM apps? | Read it for |
|---|---|---|---|---|
| Designing Machine Learning Systems | Chip Huyen | O'Reilly, 2022 | Yes | Deployment, monitoring and retraining decisions |
| Machine Learning in Production | Christian Kästner | MIT Press, 2025 | No | ML as a software engineering discipline; full text online |
| Machine Learning Production Systems | Robert Crowe and others | O'Reilly, 2024 | No | Data and model pipelines at scale |
| Reliable Machine Learning | Cathy Chen and others | O'Reilly, 2022 | Yes | Running models with SRE practice |
| Machine Learning Engineering | Andriy Burkov | True Positive, 2020 | Yes | A short overview of the whole lifecycle |
| Introducing MLOps | Mark Treveil and the Dataiku team | O'Reilly, 2020 | Yes | MLOps and governance across a company |
| LLMs in Production | Christopher Brousseau, Matthew Sharp | Manning, 2024 | No | Serving and operating language models |
| LLMOps | Abi Aryan | O'Reilly, 2025 | No | Operating practices for LLM applications |
| From Model to Product | Dulan S. Dias (this site's author) | Independently published, 2026 | No | Serving, releasing and monitoring one hybrid product |
Designing the deployment
Designing Machine Learning Systems is the best single treatment of the decisions: batch or online prediction, model compression, edge or cloud, how distribution shift shows up and what to monitor, and how continual learning and testing in production work. Read it before you choose tools.
Machine Learning in Production, by Christian Kästner, comes from a Carnegie Mellon course and treats an ML model as one component of a software product. It covers requirements, architecture, quality assurance of models and of the whole system, operations, teams and responsible engineering. Its full text is online under a Creative Commons licence, which makes it the obvious first stop if budget matters.
Pipelines and MLOps
Machine Learning Production Systems, by Robert Crowe, Hannes Hapke, Emily Caveness and Di Zhu, gives a broad view of the whole field in four parts: data, modelling, deployment, and productionising, which includes ML pipelines and pipelines for generative AI. At over 450 pages it is the deepest book here on pipelines.
Introducing MLOps is written for the organisation as much as the engineer: who owns what, how models are governed and how risk is managed as the number of models grows. Andriy Burkov's Machine Learning Engineering is the short alternative, covering the whole lifecycle in one compact volume, distributed on a read first, buy later basis.
Operating models after release
Reliable Machine Learning brings site reliability engineering to ML: service level objectives for models, monitoring and alerting, incident response, and how an organisation should run model development. Read it if you will be on call for a model, or if you lead the team that is.
For a product that mixes model types, From Model to Product covers packaging with containers, versioning models and artefacts, CI and CD, shadow, canary and blue-green releases, rollback, drift monitoring and cost, all on one running product. It is narrower than the MLOps books on large pipeline platforms. The free guides on deploying a model with FastAPI and Docker and safe rollout strategies are drawn from it.
Deploying language models
LLMs in Production, by Christopher Brousseau and Matthew Sharp, covers what changes when the model is a language model: serving large models efficiently, infrastructure, evaluation and putting LLM applications into production safely, with projects that include running a model on the edge. LLMOps: Managing Large Language Models in Production, by Abi Aryan, focuses on running LLM systems for real users: evaluation, governance, security, agents and retrieval.
If your language model is a hosted API rather than one you serve, much of the work moves to evaluation, cost and fallbacks; the guide to reliable LLM API calls and the comparison of LLM Engineer's Handbook and AI Engineering cover that side.
Which book for your situation?
| Your situation | Start with | Then |
|---|---|---|
| First model going live, small team | Designing Machine Learning Systems | From Model to Product (disclosed above) |
| Want a thorough text you can read online | Machine Learning in Production | Machine Learning Engineering |
| Many models and data pipelines | Machine Learning Production Systems | Introducing MLOps |
| On call for models in production | Reliable Machine Learning | Designing Machine Learning Systems |
| Serving your own language models | LLMs in Production | LLMOps |
Frequently asked questions
Which MLOps book should a beginner read first?
Start with Designing Machine Learning Systems, which explains why deployment and monitoring decisions matter before naming tools, or with Christian Kästner's Machine Learning in Production, whose full text is online. Move to a pipeline-heavy book such as Machine Learning Production Systems once you run more than a handful of models.
Are MLOps books from 2020 still useful?
Their principles mostly hold: versioning, testing, monitoring, governance and ownership have not changed. Their tool recommendations have, and they say little or nothing about language models. Read them for the practices and pair them with a recent book or current documentation for tools.
Is there a free book on machine learning in production?
Yes. The full text of Christian Kästner's Machine Learning in Production: From Models to Products, published by MIT Press in 2025, is online under a Creative Commons licence. Andriy Burkov's Machine Learning Engineering is distributed on a read first, buy later basis.
Which books cover deploying large language models?
LLMs in Production, by Christopher Brousseau and Matthew Sharp, covers serving and operating language models, and LLMOps, by Abi Aryan, covers operating practices for LLM applications. LLM Engineer's Handbook deploys a full LLM system on AWS as part of its project.
Which book shows how to deploy a model in code?
LLM Engineer's Handbook deploys a complete system in code on AWS. From Model to Product, by this site's author, serves models with FastAPI and packages them in containers on a running product. Designing Machine Learning Systems and Reliable Machine Learning are deliberately light on code.
Which deployment book covers edge devices?
LLMs in Production includes a project that deploys a small model to a Raspberry Pi, and Designing Machine Learning Systems discusses edge deployment and model compression. From Model to Product, by this site's author, runs an image classifier on a phone and falls back to a server model when it is unsure.
Where does From Model to Product fit among MLOps books?
From Model to Product, written by this site's author Dr. Dulan Dias, covers deployment for one product end to end: containers, versioning, CI and CD, safe releases, rollback, monitoring and cost, for a system that mixes rules, a vision model and a language model. It is narrower than the MLOps books on large pipeline platforms.
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.