AI Engineering
What is AI Engineering?
🎯 By the end of this lesson you can…
- Define what an AI Engineer actually does day to day
- Distinguish AI Engineering from ML research and data science
- Map the stack you'll learn to build in this course
AI Engineering is the discipline of building products on top of foundation models — large language models, embeddings, and multimodal models — rather than training those models from scratch.
You don't need a PhD or a GPU cluster. Modern AI Engineering is mostly about orchestration: prompting, retrieval, tool use, evaluation, and shipping reliable systems around a model you call over an API.
The AI Engineer's job
Compose, don't train
You wire together models, data, and tools into a working system.
Reliability first
You turn a probabilistic model into a dependable product with guardrails and evals.
Product sense
You decide where AI actually helps the user — and where it doesn't.
Iterate fast
Prompt → measure → refine is the core loop, measured in hours not months.
Where AI Engineering sits
How it differs from ML research
If training a model is like brewing the coffee beans, AI Engineering is like being the barista — you don't grow beans, you craft a great cup for each customer, consistently, at scale.
- ML researcher — designs and trains new model architectures.
- Data scientist — analyzes data, builds narrow predictive models.
- AI engineer — builds user-facing systems around pretrained foundation models.
Do I need to know the math behind transformers?
Not to start. You'll build far more effectively by understanding behavior — context windows, tokens, temperature, failure modes — than by deriving attention. We cover just enough theory in the Foundations section to make you dangerous.
What you'll build in this course
A RAG assistant
Answer questions over your own documents with citations.
A tool-using agent
Let the model call functions and take actions.
An eval harness
Measure quality so you can improve with confidence.
⭐ Key takeaways
- AI Engineering = building products on foundation models, not training them
- The core loop is prompt → retrieve → tool-use → evaluate → ship
- Product judgment and reliability matter as much as model choice
📝 Practice questions
In one sentence, what does an AI Engineer primarily do?
Name two things an AI Engineer does that an ML researcher typically does not.
Why is 'reliability first' emphasized for AI products?