Transcription
If you're interested in becoming an AI engineer, then watch this video. I'm going to give you a complete road map, a step-by-step guide on how to break into this field for beginners. I'm going to try not to use a ton of fancy terms. I'm going to make this very clear and easy to understand. And I'm going to explain practically how you actually could become an AI engineer in 2025 and beyond.
Okay, so let's start by defining what an AI engineer is. Now, this is actually pretty difficult. It's not something that I can give an 100% answer on and that's because this role is constantly changing. AI is evolving extremely fast and all kinds of companies have a different idea of what an AI engineer should be. So in this video I want to focus on a very practical pragmatic AI engineer which really means that you're responsible for implementing AI into software systems. It doesn't mean that you're going to build an AI model from scratch. It doesn't mean that you're going to become the next OpenAI and train something like, you know, GPT5. That's really reserved for people that are significantly smarter than probably all of us that have done massive amounts of academic research, PhDs, you name it. That's not what I'm talking about here. And that's actually really much more machine learning and like AI research than it is practical AI engineering. Most of the AI engineering roles that I'm aware of involve using existing tools, things like Langraph or Langchain, maybe doing something like fine-tuning an LLM, doing some prompt engineering work, understanding how to use these APIs, which LLM is better for a specific task. That's really what I'm talking about here. So to summarize it, it means you are actually really practically building systems that use typically existing AI models or kind of base LLMs. You may be doing something on top of that or enhancing the system. But a lot of this is actually core software engineering work that just happens to revolve around using AI, right? AI really being all of these crazy models like LLMs or existing machine learning models that people much smarter than us have already built.
So with that in mind, let's get into the road map. Now the reality here is that if you are doing this, you are going to be using Python. Python is just the number one language when it comes to machine learning and AI. It's very easy to learn. It's used in this field massively and all of the main frameworks are written in Python. So, you're going to need to learn Python. And what that means is that you need to have a really strong foundation in Python and programming in general. This means you should understand loops. You should understand variables. You should understand functions. You should know more specific Python features, even more advanced features, things like decorators, generators, potentially meta classes. You should understand packages, modules. I have all kinds of Python road maps out there. But the point is, you're probably going to have to spend at least two months here learning Python, getting good at it, and just learning standard software engineering with Python before we can dive into AI and ML. This is not a beginner field. It takes a good amount of time to get good at this. So, make sure that you have a really strong Python foundation.
Now, once you built a strong foundation in Python, there's a whole list of tools that you're going to need to learn that are very likely going to be used for an AI engineering rule. Now, the first is Git and GitHub. That's for version control and essentially saving and checkpointing your code. You're going to need to understand how to use various IDEs or editors, things like PyCharm, cursor, VS Code, etc. You're likely going to need to understand basic bash commands or terminal commands. So, how to navigate a command line interface, just something that's pretty common when it comes to being a programmer or software engineer of any type. Then you'll need to understand things like virtual environments and how you isolate dependencies for your Python projects. And then lastly, things like Jupyter notebooks so that you actually can run and experiment with different Python code.
Okay, so that's most of it. And then additionally, I will mention that having a math background or a computer science degree would obviously be helpful here. However, it is not a requirement. And in fact, most of these AI engineering roles that I've seen don't actually require a lot of advanced mathematics. Like I said, in most cases, you're actually using existing models and building systems around that as opposed to really developing the models or the LLMs on your own.
So once you've built a core foundation in Python and the core software engineering tools, the next thing that we want to move into is understanding LLMs. Now this is one of the main things that you're going to be using, especially in today's age. You know, in 6 months it might change, but for now it's LLMs. And that means that you need to understand what an LLM is, the architecture of it, how it actually works on a lower level. So you know you get some input, how do you actually generate some output and then you should be aware of a lot of the different LLMs that exist and when you should use which one. For example, why would I use Gemini over using GPT5? Why would I use Claude or why would I use a thinking or reasoning model? These are all important things that you need to understand.
Now, beyond that, you want to start learning how to use the APIs for these various LLMs. So, I'm talking about things like the OpenAI API where you're able to actually interact with GPT models and send a request to them and get a response back. Same thing like maybe with the Deepseek API or the cloud API is not too hard to do that, but it's something that we need to have a foundation and understand how to do essentially. After that, I'm going to recommend looking into tools like Olama or like Docker Model Runner. This allows you to run LLMs locally on your own computer. And again, it's important to understand how to do that. Essentially, in this section, you want to understand what LLM are. You want to start using them, and you want to start interacting with them from code.
Now, let's keep going here, but I want to share with you the reality when it comes to learning these skills. Now, AI engineering is one of the fastest growing tech roles today. There's thousands of job postings asking for real hands-on AI skills where just reading articles or watching tutorials isn't going to be enough. Now, studies show that you only retain about 20% of what you passively consume. But when you learn by doing, writing code, building projects, and getting hands-on experience, your retention can jump to 75 to 90%. Now, that's exactly why I recommend Data Camp, who's sponsoring today's video. Now, I've personally used Data Camp for years to level up my Python and AI skills. And their associate AI engineer track is built for developers who want to get into AI fast with a hands-on project focused approach. And its curriculum mirrors almost exactly what I'm sharing with you in this video. Now, with Data Camp, you'll write real Python code in your browser, get instant feedback, and build practical skills step by step. Now, here's what's inside this track. You have core AI building blocks. everything from Python foundations to APIs, vector databases, and LLM workflows. You have practical AI projects like building chatbots, summarization tools, and automation scripts that you can actually show off in a portfolio. And you have job ready skills like deployment basics, prompt engineering, and understanding how to integrate AI into real applications. Plus, there's a direct pathway to earn an industry recognized AI engineer certification that you can showcase on your resume and on LinkedIn. Now, I wish that this existed when I started out. It would have saved me months of trial and error. Instead of jumping between random tutorials, this gives you a clear guided path to become a job ready AI engineer. And right now, you can get 25% off using my link below. So, if you're serious about learning AI the right way by building, experimenting, and coding from day one, this is the best place to start and you can check it out.
Okay, now let's move on to the next section in this list, which is AI frameworks. Now, it's one thing to understand Python to be able to use LLMs, but as you start getting more advanced, you'll need to create more advanced AI systems and typically AI agents. Now, I have a few frameworks on my list that you need to be very comfortable with. Obviously, there's a lot more and you can continue to learn more frameworks, but these are the core ones for AI engineering. Now, the first one is going to be Langchain. Langchain allows you to really easily create AI applications and call into popular LLMs like GPT, Claude, etc. Now, this will let you do things like create AI agents, have tool calling capabilities, utilize vector databases, create retrieval augmented generation programs, which we'll talk about later, and it's really the starting ground for using AI a little bit more professionally.
Now, after Langchain, we have Langraph. Now, Langraph is much more complicated than Langchain. It's kind of the level up. And this allows you to do orchestration of AI agents and AI systems in general. Rather than just having kind of this passive AI agent that responds to events, you can actually have essentially a flow or a graph. That's why it's called lang graph that you can run an AI system through to get the exact desired outcome that you want. Now I'm describing this very briefly because there's a lot to go into but I would recommend start with lang chain build a few applications with that then move on to lang graph and you'll see the benefits of it and how you can create this graph and have a lot more control over what your agents and AI systems are doing.
Now after that or even kind of at the same time depends on what you're building I recommend looking at transformers and hugging face. Now, HuggingFace is kind of like an open- source platform that has a ton of different models already available for you to use. So, you can have an image classification model or a sentiment analysis model or whatever you want. It's probably already available on there and you can use it within Python by using the transformers and kind of the hugging face packages. Now, you can download the model, you can fine-tune it, you can run it exactly as it is. It's very useful and I recommend looking into those packages. Now, of course, there's other ones like NumPy and Pandas, which are always going to be good to know and a strong foundation, and some built-in modules like the OS module, the SYS module, the path liib module, but those aren't as important.
Okay, now let's move on to building projects. Once you've learned these frameworks or really while you're learning these frameworks, you should be working on building some AI projects. Now, I have three simple ideas for you. Obviously, go above and beyond this. Don't just stick with these, but I think these are at minimum what you need to build during this road map. Now, the first is going to be a simple AI to-do list. Essentially, you want to have an AI agent that will be able to create to-do list items, remove to-do list items, give you a summary, check them off, etc. Right? Pretty straightforward, but an application that you need to learn how to build. The next application I recommend is an AI web scraper. Now, I actually have many tutorials on this channel going over how to do that, but it's very interesting to see how these work. And it kind of combines some core Python skills with some AI skills, which I think is really valuable. And the last project idea I have is something like an AI content helper. So, an AI that can collaboratively help you build out content ideas, look at past experience, pull in data from something like YouTube or LinkedIn or Twitter, and give you recommendations. Now, look, there's a lot of other stuff that you can build. I'm just giving you three simple ideas that I would recommend you attempt or at least something like that so you understand kind of the core AI foundation and you know how to build kind of basic AI applications.
Now let's move on to the next topic. The next topic on my list is advanced LLM skills. Now once you know how to use the LLMs on a simple level and within some of these frameworks, you need to move on to some more advanced topics. Now I have some content on my monitor here, so I'm just going to read it off. The first thing that I have is prompt engineering. essentially how do you create the best prompts and how do you reliably get an LLM to act in the way that you want. Then we move on to fine-tuning. Now this involves taking an existing foundation model, so essentially a large model that's already been trained and passing it a smaller amount of data and fine-tuning it on that particular data or a particular task. An example of this could be taking an existing model. to something like Mistl, you know, smaller model that's open source that you could run on your own computer and passing it maybe 10,000 examples of hockey scores or hockey information and having it predict specifically based off that data. What you can do is you can essentially bias these models. You can pass them some information to fine-tune them and they can have significantly better results in a particular task with a limited amount of training data by using some of the base knowledge that they already know. That's not the best explanation of fine-tuning, but you know, hopefully you get the idea.
Now, the next thing on my list is embeddings and vector databases. Now, rag or retrieval augmented generation is a really big topic in the LLM and AI engineering space, and you need to understand what embeddings are, how you store data in, for example, a vector database, how you retrieve that data, and how retrieval augmented generation works within an LLM or within some kind of AI system. To put it simply, this means kind of extracting relevant information and passing it to an LLM to augment the information that it has so it can give you a better reply. This is extremely useful and it's one of the most practical use cases of LLMs by being able to feed it your own data or very custom data when it needs it. Next topic I have is context windows. Pretty straightforward, but you should understand the size of the output and input for these various models. why you would use a particular model, the type of data it can accept or that it can output. Moving on, we have architecture. This obviously gets a little bit more complex where we're diving into lower level concepts like how is this model actually designed? How are the layers set up? How is it actually outputting something? Um, again, this kind of goes with context windows as well and just generally the kind of layout and architecture of these more advanced models which I myself am not an expert in. Then we have things like MCP servers. How do we build an MCP server? How do we deploy an MCP server? How do we write an MCP client? How do we connect MCP servers to our AI models? Should understand everything about them because it's really the next wave of AI and what a ton of companies are pushing out right now.
Okay, that's it for that topic. Let's move on to the next one. The last topic on my list here is LLM ops or LLM operations. Now, it's one thing to have an LLM system or have some AI system running on your own computer. It's another thing to have this working in production reliably and with a bunch of users. AI in general is notorious for working sometimes but not all the times. And when we're creating software systems that are scaling to maybe millions of people, it's very important that if we are going to use AI that we build it correctly so it works reliably. So in this section I'm talking about things like Docker and Kubernetes, LLM orchestration, things like testing, things like retries, fallbacks, observability, logging, building advanced systems around these AI models or these um just general models that you have and then things like using fast API, database design, overall architecture of a program. Again, it's one thing to actually have a system that works on your local computer. It's another thing to have this working in production, be deployed, have observability into what's actually going on, have logging, and being able to write things like fast API endpoints to protect various resources to assign credits or rate limiting to users. There's so much more than simply AI. And that's why a lot of people when they get into AI engineering, they realize that it's really software engineering, but just using AI, right? Like having some AI as a component of it. And one of my friends actually is an AI engineer. He's currently working for a large startup. He makes a ton of money. And when I ask him about this, what he tells me is that his job is about 20% working with AI, right? Fine-tuning, prompt engineering, maybe, you know, pulling in some model or training it on some custom data. The other 80% is really just software engineering and building the system around these AI models to have them work reliably in production. That's the hard part. And if you can get good at that, you will be a good AI engineer.
Okay, so that is my AI engineering road map. Let me know what you think of it in the comments down below and I will see you in another video. [Music]