So, you want to become an AI engineer

but are completely overwhelmed.

What do you actually need to learn? How

hard is it? And do you need to be good

at math?

Well, let's answer those questions.

[music]

Hello everyone and welcome back to my

channel. My name is Kirk and today we're

going to go over what it takes to become

an AI engineer.

My background, I'm a software engineer.

I've been doing AI work for maybe the

last two years and I'm going to tell you

what worked for me and what I think was

going to work for you. Because a lot of

times you go online and you're going to

see a big list of all these Python

libraries you need to learn. They're

going to tell you, "Hey, you need to be

good at math and you're going to be

completely overwhelmed."

And I get it. As I said, an AI engineer

is a software engineer. That's an

important distinction.

It's not going to be easy, but if you

take this path, it's not going to be as

hard as you think. My goal is to get you

to create real programs working with LMS

and AI agents, deploy them, and solve

real business cases. I work in Azure,

but you can pick the cloud provider you

want, and I work mostly with GPT.

But again gang, we're going to go over

all that stuff, but the first thing we

should do is figure out what an AI

engineer actually is.

What is an AI engineer? An AI engineer

is a software engineer who designs,

builds, and deploys applications that

use artificial intelligence models,

machine learning, and LMS to solve real

business problems. aka you are creating

intelligent systems that perform tasks

that typically require human

intelligence.

An AI engineer builds chat bots,

recommendation systems, and agents.

One important distinction we must make

is an AI engineer is not a machine

learning engineer. A machine learning

engineer focuses on the models

themselves.

aka an AI engineer builds the full

application while the ML engineer makes

the models work. Let's look at the steps

to becoming an AI engineer.

First, the fundamentals, learning

Python. But don't be overwhelmed. You

don't have to know everything. I'm going

to tell you exactly what you need to

know.

Next, LM basics. We're going to learn

about different models. We're going to

look at prompt engineering. You're going

to need to know what the difference

between parameters and hyperparameters

are. And then we're going to look at

using LMS with rags.

Next, agents. You'll want to build

agents. That's what everyone's looking

for right now. And you're going to need

to learn some frameworks. We'll look at

Lang Chang and Langraph. And in the

Microsoft ecosystem, learn about

Microsoft agent framework.

And last, multi- aent orchestration.

You're going to need to build multiple

agents. Have them talking to each other

using this this knowledge to build real

business use cases.

First, the fundamentals.

I'm going to make some assumptions. I'm

going to assume you have strong

foundations in computer science, data

structures, and algorithms. Also, you

know about version control and

specifically Git.

Now, you're going to have to learn

Python.

No question about it. I know that

Microsoft agent framework, the new

framework is available in C, but if you

want to be taken seriously as an AI

engineer, learn Python. But don't worry,

don't be overwhelmed. I'm going to tell

you exactly what you need.

So, let's talk about some of the

fundamentals of Python that you need to

know. You're going to need to know about

the variables and types. And since you

have a strong foundation in data

structures and object oriented

programming, you're going to have to

know about classes, inheritance, and

composition.

And you're going to have to know about

functions, default arguments, variable

positional arguments, and keyword

arguments. And you have to know about

lists, tpples, dictionaries, and sets,

decorators.

and also learn about virtual

environments. You're going to probably

do some stuff in Jupyter notebooks,

especially with using hugging face,

which I'll get to in a second, but learn

about virtual environments and also

package managers. Know about pip versus

UV. I use UV a lot. A lot of the

projects I've been on, we use UV.

Let's talk about some of the libraries

that I use all the time. Pantic is a

data validation and parsing library

using Python type hints. I use it all

the time. You're probably going to use

it a lot too, especially when we get

into prompt engineering. Another thing

we're going to have to talk about and I

know you come from a background,

hopefully you come from another language

and you know about restful APIs. I'm

always building my restful APIs in

Python using fast API. Just learn fast

API seems to be the most popular one

right now. So just jump right into it

and I know we talked about the fact that

AI engineers are different than ML

engineers and data scientists and pandas

is usually is a library that data

scientists use a lot. Pen is is used for

data manipulation for and for data

scientists, but I use it a lot for

reading CSV files and working with data

frames and working with data, especially

when you talk about testing, which of

course brings me to

unit testing and I use piest. You're

going to have to learn about unit

testing. Again, if you come from another

background, another programming

language, you probably know about unit

testing. Just learn piest. You're going

to be running those. You're going to be

using tests.

Next, LLM basics.

Okay, we just covered the fundamentals.

While you're learning the fundamentals,

learning Python, you can also start

learning the basics of LLMs.

And while you're learning about LMS, aka

learning about different models, how

inference works, you should also start

writing code.

Sounds scary, right?

Well, it isn't.

Why? Well, Hugging Face is a great place

to start.

Hugging Face is a company and open-

source platform for AI, NLP, and

multimodal models.

Think of it as GitHub plus app store for

AI models. Let's take a look at it right

now. Here we are on huggingface.co.

Now, here is where you can go and learn

about NLPs

and LLMs, the very basics. This is what

I did. I went to the LM course first and

I stepped through this and as I said, I

didn't say you need to know about

PyTorch or TensorFlow and Python

libraries because you're going to be

exposed to that a bit and you're going

to be working with Jupyter notebooks.

But this is a great way to learn about

transformers,

tokenizers,

and inference.

great foundations that you need to know.

I went through this course, great

course,

all the fundamentals. I don't do a lot

of this on my job, but you need to know

this stuff. And then after that,

I also did the where is it here?

I also did the agents course which is

going to get you exposed to different

agent frameworks specifically the one I

use the most now is lane chain and lane

graph and of course in the Microsoft

ecosystem Microsoft agent framework and

using LMS which I'll talk about in a

second as well but this is a great way

to learn about frameworks what's

available lin index small agents and

building a gentic Greg use case. Have

some fun. Play with this. Go through

this course. Get the fundamentals of

what agent frameworks and what agents

are. I should also state on the LM

course. You're going to also learn about

fine-tuning models. Again, the

fundamentals that you need. I don't do

this in my day-to-day job, but you

should know about how to fine-tune

models, be exposed. Another thing you're

going to learn with this LM course is

you're going to learn about models and

NL NLP tasks, specifically

summarization classification text

generation, and you're going to be

fine-tuning models. You're going to know

the difference between what a

pre-trained model is, a fine-tuned model

is. Again, I don't do a lot of this. I

know I have a tutorial on fine-tuning

models.

the foundations that you need to know.

Again, I'm gonna always say that. But

you're going to be exposed to that.

You're also going to learn about

hyperparameters and the difference

between hyperparameters and parameters.

Parameters being the size of the model

like 1.5 billion, 10 billion. Again,

this is stuff like you're going to learn

about open source models. And then of

course once you get in the open AI in

the Microsoft world you're going to know

about what

GPT is which is a prop proprietary

model. If you go back to the homepage

and click on models you're going to be

exposed to different different models

here classifying by the task by the size

and then looking at the model cards

themselves and not just LMS but

multimodal models as well.

Great place to get read up about models,

learning about different LMS. Hugging

face is your friend. Play with this.

Now we need to talk about prompt

engineering. Prompt engineering is the

practice of designing and refining

inputs aka prompts you give an AI model

so produces more accurate, useful, and

predictable outputs.

A lot of my time as an AI engineer is

just spent writing and editing prompts.

I work with business analysts, with

product managers, figure out the

business use cases, write some PC's,

and show them what we can do, and then I

just fine-tune the prompts.

Yep.

I'm just writing language, not code. But

that's part of being an AI engineer. But

this is really important. So get good at

it and get used to it. And that brings

us to our next topic.

Next, agents and multi- aent

orchestration.

Okay. Next, let's talk about agents and

agent orchestration.

First,

lang chain.

Okay. Let's talk about lang chain. Now I

know in my channel I'm always talking

about Microsoft and staying in the

Microsoft ecosystem and introduce

introducing you to Microsoft agent

framework. But for most of my projects

and to be an AI engineer, you should or

you have to know lang chain lane chain

and lane graph. And the best way to

learn that is just go to lanechain.com

which I'm at right now. And you see at

the top here, learn and you go to the

lane chain academy.

And I went through all this. I did this

a while ago. I think in 2024, so a year

and a half ago from this recording, but

this is a great resource. And I did the

foundations which you start building

agents. And I also learned about

Langsmith

which is about observability and seeing

how these models and these agents work

and how they reason.

Great tools. Learn this stuff. Of

course, if you're in the Microsoft

ecosystem, they also have very similar

tools for observerability, which we'll

talk about next. But just learn this

stuff. Learn lang chain, learn lang

graph, master this. And this, as I said,

is a great resource.

Um, click on here, you can go through

it. The only thing about this is you're

going to be in Jupyter notebooks a lot.

And again, as an a engineer, you want to

get out of Jupyter notebooks. if you

want to start building real

applications,

which we do using my tutorials, which we

will get to next.

And that brings me to YouTube and my

channel. Yes, maybe I'm biased, but I

think a great way to be an AI to become

an AI engineer is to go through some of

my content here. But where do you start?

I have a couple playlists here. The best

way to learn about the fundamentals of

Azure and what and Azure AI foundry but

now Microsoft Foundry is to go through

this playlist here. Let's just look at

the playlist.

I would go through here and step through

all the videos. Maybe you already done

that. Of course, you subscribed. You

probably have. But learning about LM

with rags at the start and learning

about well fine-tuning as I said I don't

really do my day-to-day job but you're

going to learn about agentic rag and

everything else that this has to offer

in terms of debugging your agents guard

rails. You're going to learn about how

to evaluate different models, what

models are available in Microsoft

Foundry and building multimodal and then

and then you're going to be using the

guey itself and learn about Microsoft

Foundry and workflows.

After that, you're going to want to know

about the different frameworks available

to building agents in Python. Now we

have Microsoft agent framework used to

be semantic kernel and autogen. I again

have a playlist here for that. Let's

look at the playlist itself.

So we I would just step through this

again. Maybe you already have but you

want to learn a framework. You want to

learn Microsoft agent framework. Okay

that covers it for what it takes to

become an AI engineer.

But as I said at the very start, an AI

engineer is also a software engineer. So

it's great to learn this stuff that I

just talked about, but you also need to

know how to be a great software

engineer.

Become a platform engineer.

Learn the cloud. Pick your cloud

provider of choice. For me, it's Azure.

And really get good at it.

learn what rate limiters are, API

gateways,

and also

we are all people, right? We sometimes

forget that. As I said when I talked

about prompt engineering, a lot of my

day is spent working with product

managers and business analysts figure

out what we actually need to build and

editing those prompts to create a real

product. something you can be really

proud of. And I hope you follow these

steps and take my advice, you'll really

get something out of it.

And if you did get something out of it,

also remember to please subscribe

[music] and like this video. And until

next time, I'll see you then.