Transcription
If you're looking to become an AI engineer, if this job is on your radar, then this video is all you need. It's a full breakdown of what this role is, how it differs from software engineering and machine learning engineering, um how to become one. So, I'm going to give you a full road map, what projects you should be doing, what kind of person this job would suit and also my opinion on where I see this role going and is is it worth your time to invest in.
Okay, before we go any further, I got into tech in my 30s. I learned from scratch all this stuff and I watched loads of videos like this. I'm not just going to give you a random road map you can get off track. when I was learning, I wish someone taught me this stuff, but also said, you can go out there whilst you're learning and build your own products and sell them out there in the world. So, you can learn to become an AI engineer and you can also make a SAS, make something, sell it. Worst case scenario, it looks good at an interview. Best case scenario, it becomes a business.
So, I'm here my page, everything on my content is all about encouraging you guys to learn technology, but also believe in yourself, sell your own stuff, and see where it goes. Because taking a chance might lead somewhere. So, this video is going to be four steps. Each step is going to gradually increment. The end of it, I'm going to encourage you to make your own microsass. And if you're new here, my name is Andrew. I'm a senior software engineer, and I got into tech in my 30s, totally from scratch. and I have worked in the AI space as an engineer. So, I've watched this role evolve over the last few years. I make tech content around your career around building apps. So, if that sounds like your kind of vibe, subscribe if you want.
Okay, before we go though, I want to just manage expectations because firstly, I want to give you correct factual information on this channel. The chances of you getting your first role working for OpenAI are very, very low unless you look at further education. So, getting a masters or a PhD. There are a lot of jobs below that. There are so many opportunities in AI below that. So that should be your focus. So clearly the reason this is taking off and the reason the salaries are so high is that companies and the world is crying out for people who can implement AI.
Firstly though, what is an AI engineer? This is someone who builds production ready systems using pre-trained AI models and APIs. And to further set the scene here, let's talk about three roles. So there's a software engineer, there's an AI engineer, and there's a machine learning engineer. People confuse these last two a lot. So a software engineer will write code or get AI to write the code nowadays, and that will produce a program. Machine learning engineers, they take loads of data, they clean the data, they fine-tune the model to create a program or a model that will achieve some output. The AI engineer makes that model useful in the real world. Well, they'll make it into a real system or a real app.
Okay, so this is not a machine learning engineer. This is not a machine learning researcher. These are people deep in math. People who write research papers, they train the models themselves. This is not that. This is more like the practical use of AI. So, it's what comes after. And if it's still not quite clear, there's a really good article called the rise of the AI engineer where they basically say that in the future there's going to be more AI engineers than ML engineers. There's a really good visual and it shows the line of an API. On the left side, you've got the data, the research, the data science side. On the right side, this is where the AI engineer comes in. So, this is the product and user focus. You got full stack engineers there. So, it's all about chains, agents, and making things basically.
With that said, let's get into the road map. So, I've broken it down into four steps. And how you learn this stuff, you learn by doing. You don't learn by watching tutorials or reading documentation. You learn by becoming a problem solver. So it's going to be four steps. At the end of each step, there'll be a project. So this is all project based learning.
Step one is to focus on the core skills of AI engineering. So Python, LLMs, and system thinking. And to start with, everyone starts with Python. This is the language of choice for AI and machine learning. The community has chosen it. You have R as well, but you have a lot of things built on top of Python or the libraries which just make it easier. For example, PyTorch. And if you've never coded or you never learned a programming language before, these are some things which help me. And the good mindset is that you're going to be learning Python for the rest of your career probably. It's not something you learn in like a year. It's just a continuous process. But the real thing you're learning is problem solving. And this is hard to teach. What helped me was doing toy problems. So there's a site called Code Wars. It's totally free. Start right at the beginning. And whenever you're learning, try and start every session by solving one of these small problems in Python. Also, only watch one video tutorial on Python and then build and use documentation. There's a few good courses. You could do YouTube courses, loads of free Code Camp ones. There's a bro code one if you want a paid one. And the one I learned from, it's zero to mastery. They got a really good structure, really good like projects and that's how I learned Python myself. But the mindset is becoming a problem solver. So you want to get them comfortable and learn all the basics of Python and also Git and GitHub. So what you want to do is these projects I suggest you want to be committing them to Git. So you don't have to know everything about Git. It's just the main things like pushing, pulling, and merging. There's a site called Learn Git Branching which is quite good. There's a few YouTube tutorials you can watch and also a more advanced book called Pro Git. So these are all free resources, but again the kind of thing learn it early on because you want to be committing all your stuff because this is the workflow that real engineers use. So learn Git as soon as you can.
Okay, next we're actually getting into the real AI engineering. So we're going to call and go through OpenAI's API documentation. There's loads there. They're the biggest provider, so it's a good start for you. This is like one of the key parts of being an engineer. So calling APIs, doing orth requests, how to handle data, what data you get back. There's actually a fair amount involved with like gluing APIs and it's something that took me a while to get to grips with. So yeah, OpenAI is a great start. They've got loads of documentation. So get stuck in with the Python part.
Okay, the next part of the fundamentals is learning about the LM basics and prompt engineering. So this is something that all software engineers are having to learn right now. Basically, how to use an LM, how to use AI effectively. So, the context window, temperature, tokens, input and output patterns, learning all this stuff, you'll need it in this role. Thank you to Data Camp for sponsoring this part of the video. Now, I've said this loads of times, but I taught myself to code and got into tech at 30 and investing in my learning of some of these tech skills was without doubt the best return on investment of my whole life. And if you are serious about becoming an AI engineer this year, the fastest path is not just binge watching tutorials, it's building real systems. This is why I recommend Data Camp because it's really hands-on. They've got short lessons, in browser coding, guiding projects, and instant feedback. And for most of you watching, you want to look at this one here. It is the associate AI engineer for developers track. Why? Because it uses the exact stack that you'll use to ship AI features like the OpenAI API, hugging face, lang prompt engineering, embeddings, and LLMs. But you're learning how to build chat bots, semantic search, and productionready AI apps, not just playing with prompts. And if you're coming from a data background, then data camp also has this the associate AI engineer for data scientist track. And this one's interesting because it goes deeper into PyTorch fine-tuning models like Llama 3, MLOps, and deployment. And what's really cool is that these tracks prepare you for industry recognized data camp AI engineer certifications. So basically, you're getting the best of both worlds. Practical experience, but also a pathway to earning these industry recognized certifications in one place. So if you want one structured practical road map instead of just guessing what to learn next check out data camp using the link in the description and wherever you're starting from there will be a track for you and if you do check it out hope you enjoy it.
All right and the last part of the fundamentals is to start thinking in terms of systems. What do I mean by this? So think of a small AI workflow and just kind of sketch it out so you have an idea of how all the pieces fit together in terms of like a big picture bird's eye view or like a tiny agent framework. So just sketch it out just so you have an understanding of the big picture. This will really help you going forwards.
For the final step, I mentioned we're going to do a project which is going to bring everything together that you've learned and it's going to be a personal research assistant CLI using Python and also calling the OpenAI API. So basically, you'll ask it questions and it will structure the answers, but also you can ask follow-up questions. So this is a nice project. You're going to use Git, you're going to use Python, you're going to call the OpenAI API and put it all together that you've learned.
Quick pause. If you're enjoying this content, if you're getting any value from it, it's totally free for you. I would massively appreciate it if you like the video and subscribe if you want. It would help me out massively. It motivates me to make more videos like this to encourage you guys to build, to learn, and to use technology to change your life as it has done for me. Back to the video.
Okay, step two. Now we're getting into the nittygritty, the real stuff. So, backend architecture and rag. So we're going to learn how to build real backends with fast API and pedantic. So if you're a software engineer now, maybe you know this already and you can just add on these new technologies or if you're a computer science student, maybe you use Java, whatever, then just shifting learning backend but learning pedantic and fast API and in particularly things like async endpoints, uh background jobs, dockeriz services, kind of key fundamental parts of backend working as an AI engineer. We're going to pick up Postgress. So Postgress really popular. I learned it a couple years ago. There's loads of resources on YouTube. You won't struggle to find resources for free on this. We're going to look at cues, migrations, event- driven patterns, and just getting to grips with a database technology. You can actually build a decent amount of stuff with just step one. But step two is all about real backend services. So not local. All right? Because when you go into a real job, they're going to be using a lot of different things in the environment. They're going to have a database. they're going to use docker probably and obviously uh backend like fast API. So getting really deep into this and I spend a lot of time in this area and after Postgress at this point in step two we're going to learn rag learn rag end to end fundamental concept going to using it a lot so embeddings uh vector databases chunking all this kind of stuff there is a really good 2hour I think free code camp tutorial completely free I used a year or two ago it's really good and the whole goal of section two is to design and implement an AI backend using fast API that talks to a database using Postgress and has custom data sources.
So the project, so every step will have a project. For this one, it's going to be a docs Q&A back end. So it use a fast API service where you can basically upload PDFs and markdown notes. Then we'll chunk, embed, and store in a vector database. So that will use Docker, which again you can use free code camp to learn. It's one of the most used technologies in tech. You will not struggle to learn it. and also a Postgress database.
Okay, next we're on to step three and we're getting deep into this learning actual real skills which are valuable in production ready AI systems like monitoring eval and safety because AI systems can be tricky. This is not just step-by-step programming. They can hallucinate. They can just add in randomness. So, we need a way to measure the cost, the efficiency and reliability of our AI systems. So, this is what we're doing now.
Okay. First, we add in observability and tracing. And there's a tool called Langfuse for this. It's open- source. And what that will do is that every LM call we make, it will log the inputs, outputs, the latency, and the cost. So, we can evaluate as we go on, we can see how effective it is and yeah, try and reduce that randomness and those hallucinations. Next is the safety element. So, we add in guard rails. for example, prompt injection defenses so hackers can't trick a bot or filters to catch and redact personal information, output validation to make sure responses follow our format, and also safety rules to block harmful content. So there's a tool called Sentry as well which can catch those runtime crashes. This is key because this is what differentiates real AI engineers who work with production ready apps. So if you can talk about this in an interview, the goal is to measure the quality of our system with numbers so we can catch regressions before a user would notice and just keep the whole system reliable and safe.
Okay. And the project for step three, we're just going to take what we made in step two, the docs Q&A, and just make it production ready. Add in what we've learned in this step. So we're going to layer on Langfuse tracing so we can see every LM call. We're going to add in the test data set. So just add in like 50 or so real questions. We're going to hook up LM's judge scoring. So GPT4 will automatically just grade the responses. Add in the stuff we learned. So prompt injection filters, redacting personal data and add in Sentry as well. So we can just catch any crashes we get. And boom, that is a production ready app. And that will look quite good at interview.
Okay. And step four is deployment. So getting this real production ready AI system that we've built and getting out there in the wild. And there's a lot to this. It's CI/CD, which you you're going to have to learn. It's cloud. It's just all the little things involved with making something locally and then shipping it, getting out there in the world, making it reliable, making it the infrastructure good. Let's get into the steps. And how we do this is first you want to pick a cloud provider. So I recommend AWS. You can also do Azor. learn about the platforms, you know, whichever one you choose. This is going to be a key part of the role. Obviously, if you're a back-end developer, you probably know about this already. Then get your Dockerized backend running there with proper HTTPS, environmental variables, and also logging set up. Next, you want to get into CI/CD. Most companies, most web apps nowadays are set up with this kind of pipeline unless they're really backwards. And it will automatically deploy without you having to do anything. So you can also add in monitoring, health check endpoints, alerts when things break and also just to keep an eye on costs. The goal with this step overall is to show an employer an interview that you can independently build, deploy and run a production AI application without needing your handholding. So the way you do this is just ship a few of them. I recommend three. And yeah, get this out there. Put it in a GitHub repo. Put it out there for people to see. Showcase yourself. show of everything you've learned. So, LMS, APIs, rag, you know, a good clean backend structure with monitoring. It's all about showing yourself off, showing good code, and applying for jobs.
Okay. To finish, we're going to make the last project on step four, and we're going to bring it all together. And I think if you can go to an interview and show off this project, it looks pretty cool. So, it's going to be a production ready AI Micros. What we're going to do is take the Docs QA bot we made, add on a simple web UI so people can see in a browser. So just use something like Gemini, which is good for front ends. That's good because it'll teach you more about front end as well. We're going to add in user authentication so it's not wide open. A pricing gate with a free tier but also paid API limits too. Then deploy it all with CI/CD which we learned earlier with health checks, some basic metric dashboard, a public landing page as well with documentation. So you can go to an interview and this is a real product but also you can maybe even charge for it yourself. So when you're making this think of something that you actually want to build something which has like a production AI microsass and just try and build it.
Okay. And that's the steps. So now you're in the position where you can apply for an AI engineer role but also you can actually build your own AI microsass. So hopefully this video helped. I didn't want to just make a random video of a road map. CHBT can provide that for you. But if you go through these steps and appreciate that this is not just like you don't learn Python and then move on. You're constantly learning these things and if you keep improving someone will take a chance on you. You might have to be pragmatic with your first job. I was I started off as a WordPress developer then was a software engineer in an AI startup. So hope that helps. If you got value from this video only thing I ask I appreciate it if you like the video and subscribe if you want. Happy coding. I'll see you in the next one. Ciao.