About These Notes
Welcome! These are my personal learning notes from reading Chip Huyen's AI Engineering: Building Applications with Foundation Models (O'Reilly, 2025). I wrote them to capture what I learned in plain English, in my own words, and I'm sharing them in case they help others on the same journey.
Why this book matters
When ChatGPT arrived, it didn't just impress people, it lowered the barrier to entry for building AI products. You no longer need a research lab and a fleet of GPUs. You can build a useful AI application by calling a model someone else trained.
Chip Huyen calls this new discipline AI engineering: the process of building applications on top of readily available foundation models. It grew out of traditional ML engineering, but the scale and capabilities of modern models create new opportunities and new challenges.
In my notes I tried to keep the book's core promise: focus on fundamentals that last, not tools that change every month.
What "AI engineering" actually means
The one-sentence definition
AI engineering = adapting powerful, pre-built foundation models to solve real-world problems: through evaluation, prompting, context (RAG/agents), finetuning, data work, and inference optimization.
The key shift from traditional ML:
| Traditional ML Engineering | AI Engineering |
|---|---|
| You train a model from your own data | You adapt a model someone else trained |
| Heavy on feature engineering & labeling | Heavy on prompting & context construction |
| Model is the hard part | Evaluation and product are the hard parts |
| Smaller, task-specific models | Large, general-purpose foundation models |
The questions this book answers
- Should I even build this AI application?
- How do I evaluate it? Can I use AI to evaluate AI?
- What causes hallucinations, and how do I reduce them?
- What are the best practices for prompt engineering?
- Why does RAG work? What strategies exist?
- What's an agent, and how do I build and evaluate one?
- When should I finetune, and when should I not?
- How much data do I need, and how do I check its quality?
- How do I make my model faster, cheaper, and more secure?
- How do I build a feedback loop to keep improving?
How these notes are organized
The book follows the natural order of building an AI application, and so do my notes:
- Foundations (Ch 1–2): what foundation models are and how they work.
- Evaluation (Ch 3–4): how to know if your system is good.
- Improving quality (Ch 5–8): prompting, context, finetuning, and data.
- Production (Ch 9–10): making it fast/cheap, and the full architecture + feedback.
Each chapter page has the same shape:
- In one minute: the gist.
- Key ideas: the concepts explained simply.
- Decisions & trade-offs: what to actually do.
- Takeaways: what to remember.
Ready? Start with The Big Picture.