Prompting and context
Specify behavior, boundaries, and useful inputs.
Coming soon
10,000+Copies sold
4.5 ★213 Amazon ratings
You might already have the first one.
Building LLMs for Production got readers to their first working LLM application. This one carries that foundation forward and builds on it: harnesses, deployment, evaluation, monitoring.
Building LLMs for Production, by Louis-François Bouchard and Louie PetersA preview of the ideas covered
Reliability is built around the model, layer by layer.
Coming soonA reliability system, layer by layer
Retrieval and knowledge
Ground decisions in relevant evidence.
Workflows and agents
Coordinate tools, state, and approvals.
Evaluation and observability
Measure behavior and inspect failures.
Harnesses and recovery
Control retries, fallbacks, and improvement.
Model capability is the foundationThe complete system creates reliability.
What changes for you
From “it worked once” to a system your team can test, observe, and improve.
Not a tour of one framework. The book focuses on the decisions and failure modes that outlive whichever one you are using.
Know why it failed
Traces and failure categories turn a bad answer into a specific thing you can fix.
Change a prompt and know if it helped
Representative cases and graders that tell you whether last week's change actually helped.
Keep the agent inside its limits
Permissions, checkpoints, and human approval placed where being wrong costs the most.
Get the right context in front of the model
Ingestion, reranking, and provenance built around what the system actually needs to know.
Hand it to your team
Monitoring and operating patterns so the people who did not build it can still run it.
Who we wrote it for
You can vibe code an app in a weekend. Keeping it working is the skill.
Some coding background helps, but you don't need to be a researcher or a senior engineer. The book takes you through the full AI engineering stack from the beginning.
You want the full stack, not another tutorial.
You've used model APIs or a framework, and now you want the engineering underneath them.
You've shipped something and it mostly works.
Retrieval that misses, outputs that drift, tools that misfire, and no clear way to tell why.
You need it to hold up without you.
Evaluation, monitoring, and operating patterns a team can run after launch.
Why we wrote this book
We set out to update the first book.
The funny thing about writing a technical book is that the industry sometimes changes faster than the book does.
When we published Building LLMs for Production in 2024, the problems people were solving looked very different. Two years later we went back to ask what needed updating.
Almost the entire conversation had moved. Not just better models. New tooling, new architectures, new failure modes, new ways of evaluating and deploying these systems.
So we didn't write a second edition. We wrote a new book.
Louis-François and Louie
Meet the authors
Two authors. Two halves of the problem.
Louis-François Bouchard
Louie Peters
Join us on launch day
What to expect on October 20.
Before the book
Louis-François walks through his own AI engineering setup, and how to improve yours before you start.
What's inside
What we covered, why we structured it this way, and where we'd tell you to start.
Questions
Time with Louis-François, Louie, and the Towards AI team.
The full running order is still being finalised. Register and we'll send it as soon as it's set.
Join us on October 20
Register for the launch.
- Confirmed session times
- Access details
- Book updates
- Launch reminder
We'll send confirmed session times, access details, and occasional book updates. Nothing else.