Compare Machine Learning and AI Engineering

0:00Hello and welcome to the next skill in the introduction to machine learning and AI engineering

0:06course.

0:07And in this skill, we're going to compare machine learning and AI engineering because

0:12well, this is a course titled on both of those domains.

0:16And before we go deeper into models, tools, frameworks, code, we really need to separate

0:22these two core domains because right now a lot of people still use the word AI to mean

0:28almost anything like machine learning, deep learning, chatbots, prompting, automation,

0:33even analytics.

0:34Oh yeah.

0:35And agents, all of it.

0:37And that's going to make things a little confusing, especially in the AI engineering part.

0:42We did a pretty good job of defining AI, machine learning, deep learning, and generative AI,

0:47but AI engineering, not so much.

0:49For example, if we're training a binary classifier and we did that already, and that is hot dog

0:55and hamburger, and we use teachable machine to do that with no code, we are essentially

1:00training that classifier on labeled data.

1:05And by labeled data, I mean hot dog, hamburger.

1:08We even had another class, but let's just think of it as just two for right now.

1:12We added nothing in the previous skill because it's a lot neater if you don't have an object

1:17and it says nothing.

1:18Right.

1:19I did that just to make it very clear, but in this case, it's binary.

1:23It's only going to predict hot dog or hamburger if it meets a certain threshold.

1:30That way we don't have to have that third nothing class.

1:33So this is not the same thing as number two, which would be building a document assistant

1:40that's built on top of a foundation model and foundation model.

1:44That's going to be things like Grok, Claude, ChatGPT.

1:48These are models that are also called large language models, trained on large amounts

1:52of public data like the internet so that it has a foundational knowledge as a general

1:58model.

1:59And you can build a document assistant using something called RAG, which we're going to

2:03talk about later.

2:04But in essence, retrieval augmented generation means that you're going to give it information

2:09or a document so that this foundation model can help you with that document.

2:14You don't need that document if you don't need any outside information.

2:18For example, if it's contained inside of this foundation model, you're good to go.

2:22If you need some proprietary data, you can use RAG.

2:26So number one, you're training a model on label data.

2:30Number two, you're using this existing LLM or foundational model to build some kind of

2:35an application to help you with documents or to upload and then help you with that specific

2:43document.

2:44These are two different workflows.

2:46So a person can absolutely work with AI and barely train any models.

2:51And that is a very common confusion.

2:53Also, a person can absolutely build models and barely touch prompt engineering.

2:58So you could do one or two and never touch the other.

3:02So one is a machine learning engineer.

3:04That is number one.

3:05And number two is an AI engineer.

3:08And the takeaway is that these are different modes of work.

3:12So we're going to clearly define machine learning and AI engineering in this context

3:16in this skill.

3:18But we're also going to talk about data science, data analytics, because they do overlap, especially

3:25data cleaning and where that fits into the picture.

3:28Both data scientists and data analysts have to clean data if they're going to do work.

3:33Also, data science includes machine learning engineers.

3:36If you're a data scientist and you can train these models, well, you have overlapping domains.

3:41So we're going to clear that up as well.

3:43And best of all, we're going to make all of this familiar using spam classification.

3:49So no more hot dog hamburger.

3:51Now we're going to think about emails, right, which is a very common task.

3:56And we're going to look at it in three ways using software 1.0, handwritten rules, software

4:022.0, a learned model, and software 3.0, which is prompt based interaction, which veers into

4:08AI engineering.

4:10So the goal here is to help you build that mental model because the course gets much

4:15easier to follow and understand when you have a clear mental model.

4:20All right.

4:21So what are we going to cover in this skill first?

4:23And also, there's a very specific order here.

4:26We're going to define machine learning in a deeper, more practical way.

4:30Second, we're going to go right into AI engineering.

4:33This is the first time we're focusing on AI engineering and not generative AI.

4:38And we're going to do this in the same way we do machine learning.

4:41We'll keep it broad enough, though, to be accurate, but very granular and clear enough

4:47that you can immediately tell how they differ from machine learning and critical areas that

4:52are going to apply to the rest of the course.

4:54And third, we're going to get into data science, data analysis, and data cleaning.

5:00Data cleaning, because we're in the age of AI hype, people love to talk about models

5:05or prompts, but they somehow ignore the condition of the data.

5:10And that is a catastrophic mistake.

5:13I can't tell you how many times I've seen this and heard about this because these chat

5:18bots can spit out code, advice, and two key things happen that are really dangerous.

5:25One, your knowledge is as far as you can go with the chat bot.

5:29It meets you where you are.

5:30And if you don't have a data science background, you're really not going to worry about data

5:34cleaning.

5:36So nobody really does.

5:37And if you have bad data, it doesn't matter how well you train a model or how well you

5:43engineer a product, garbage in, garbage out.

5:46So cleaning data, getting the right kinds of data, and putting them into the model or

5:52to the AI engineered application is just as important.

5:55And that's why we're going to talk about this, because it's probably the most important and

5:59overlooked part of machine learning and AI engineering.

6:03And once we finish with that important topic of data science, data analysis, and more importantly,

6:09data cleaning, we're going to do a quick callback on software 1.0, 2.0, and 3.0, and we're going

6:17to start using it because that ties directly into that mental model.

6:21You know the differences.

6:22One is deterministic.

6:24The other two are not deterministic, especially number three.

6:28And this really helps you think about how am I going to solve a problem with these different

6:32types of software or different subsets of AI.

6:37Then we're going to talk about pseudocode.

6:40And if you have a programming background, you know what pseudocode is.

6:43Let's say that I write a simple program with Python, hello world, that is probably the

6:49easiest one.

6:50And you could write the steps in English and then write the code.

6:54And the reason that that is popular in programming is it gets you to think about the logic before

6:59you write any code.

7:01It's easier to do that because you have an understanding of what it is that you want

7:04to do.

7:05And when you go to execute that and you have missing knowledge, well, you can focus on

7:08syntax and not logic and syntax at the same time.

7:12This is even more important for machine learning and AI engineering because if you don't know

7:18what the steps are in plain English, then forget about writing code.

7:23Even if you use an AI chat bot to spit out some code, again, if you don't have the domain

7:29knowledge, it's going to steer you into weird ways and dead ends and maybe even dangerous

7:36security vulnerabilities that you are going to be unaware of.

7:40So depending on your background, we're going to make sure that this is super clear.

7:44And then finally, we'll have a challenge, which is going to be really exciting.

7:48We're going to create three spam classifiers.

7:51All right.

7:52On that note, I will see you in this very first video on machine learning, and we're

7:57going to focus on the machine learning pipeline at a high level.

8:00All right.

0:00Welcome back.

0:01In this video, it's all about machine learning pipelines,

0:04really thinking about them at a high level

0:06because we don't want to get into machine learning

0:08and compare it to AI engineering

0:11without that mental model for machine learning.

0:13And the best way to do this is the pipeline.

0:15And the first one is data ingestion.

0:19This is where you take raw data

0:21that you're going to use for your machine learning project.

0:24And in any real project, these stages work back to back.

0:28So the next step is data prep.

0:30So data prep is really focusing on access.

0:33Can I ingest this data?

0:35Do I have access to this data?

0:36So you have the raw data.

0:38Now you need access to that data.

0:40And also quality is important.

0:42So this is very common in data science.

0:45In fact, a lot of data scientists will say

0:47that most of the time was spent on data cleaning.

0:50So imagine the time savings and the efficiency boost

0:53that they're going to get using AI as a force multiplier,

0:57especially for data cleaning, and also efficiency.

1:00Now notice that data ingestion and data preparation,

1:04this is where you collect the raw data

1:06and start cleaning it and preparing it for modeling.

1:08What do you think goes next?

1:10I said it already, modeling.

1:12All right, so modeling is where you actually build a model.

1:16And we'll use either PyTorch or TensorFlow

1:19later in the course.

1:20And what happens after modeling?

1:22Again, this is software 2.0,

1:24which can be deep learning, machine learning,

1:26or deep learning.

1:28And if you said training, that is correct

1:30because we already experienced the hamburger

1:34and hot dog binary classifier.

1:37And what did we do?

1:38We collected that data.

1:40Here they are.

1:41I have our friends, hot dog and hamburger.

1:43So it was really easy to collect that raw data

1:46because we created it.

1:47And then we didn't really clean it

1:49because there was no pictures of bananas

1:52or French fries or something like that.

1:54But we did make sure that it was 100 samples of each.

1:57And that has to do with quality and efficiency.

1:59So it was a very simplified version

2:02of what you would experience.

2:03But then we got into modeling, right?

2:05The modeling was done by a teachable machine.

2:08We really didn't have to code anything.

2:10It's a no-code example.

2:11But what we did do is train the model.

2:14So it went through, I believe, 50 epochs.

2:17That's 50 times.

2:19And each time it got better and better and better,

2:21which simulates the learning part.

2:24And here we measure, diagnose, and adjust.

2:27In our case, it was pretty easy.

2:29We just clicked on training.

2:30We didn't tell it how many times to train.

2:32We didn't work with all the other hyperparameters.

2:36So all you have to remember

2:37is that we have to get the raw data

2:39and prepare it for our model.

2:41We pick the model that we're gonna use

2:43and the frameworks and all of that.

2:45And we kind of build it out.

2:46And then you connect the data

2:48and start training on that data.

2:50And next is very important.

2:52This is evaluation.

2:54And here we would have a test dataset

2:57and sometimes a validation dataset.

2:59And this is how you measure the performance.

3:02So here we have a train dataset.

3:05So this is how the model learns.

3:07And then you have a test dataset

3:09with unseen values or data.

3:12That way you can compare

3:13how far off are we between these two.

3:16And that is really evaluation.

3:18And then deployment.

3:19We deploy our model.

3:21And this is where you serve that model to your users.

3:25But you also wanna monitor

3:26because you don't wanna just deploy and forget.

3:29There's all kinds of things that can happen.

3:31And without monitoring, wow, that would be a lot of risk.

3:35Now let's consider this whole pipeline

3:38because these stages don't live in isolation, right?

3:41They influence each other.

3:43For example, if the data ingestion is messy at this stage,

3:48well, it's gonna affect modeling

3:50and it's going to really impact it negatively.

3:53Also for data preparation,

3:55if we don't do a good job here,

3:57even the best models are going to perform poorly.

4:00And I've seen this way too many times.

4:03And once you don't evaluate it correctly,

4:05then you're gonna think that,

4:06hey, I have a really awesome model.

4:08I'm getting 100% on everything.

4:10But if you test it with a test dataset

4:14or validation dataset,

4:15you can see that, well, I'm actually not doing so good.

4:20You'll see a percent, maybe 100% on the training data

4:23and then 50% on the test data.

4:26And this is called overfitting.

4:28It's a common problem.

4:29If you don't validate your model,

4:31you're gonna think that it's awesome

4:32and it's completely wrong.

4:34So evaluation is critical.

4:36And finally, if we don't do a good job deploying the model,

4:39then even the best model may never be useful

4:42in the real world.

4:43But more importantly, if you're not monitoring,

4:46let's say that you had a model that was 85%.

4:49Well, if you don't monitor,

4:50you might miss the fact that it's going to 80%

4:53and then 50%.

4:55The performance is drifting.

4:56And how would you know if you're not monitoring?

4:58So these all work together.

5:00A good performant model that stays performant

5:04ensures that each one of these steps are done properly.

5:07Let's say you build an awesome model.

5:09You have great training data and evaluation and deployment,

5:13but you didn't clean the data.

5:15So your model is overfitting

5:18and then the evaluation is really where you see,

5:21wow, this model is not performing very well.

5:23And it has to do with the first step.

5:25Those two purple steps are very important.

5:28When we were working with Teachable Machine,

5:30we took pictures with a very simple user interface

5:34by holding up an object and then boom, it's working.

5:37Well, that was a very good example

5:40to understand how the process works,

5:41but that is not how we do it

5:43when we're building our models.

5:44And that's why we looked at this pipeline

5:47so you could see the importance of each stage.

5:49Most of our early focus in this course

5:51will be on data preparation.

5:53Also building models or modeling, and then training

5:58so that our models can learn from the data

6:00and then making sure that there is no problem.

6:02So evaluation, and you might be asking yourself,

6:05well, why are we skipping the other stuff?

6:07Here's why.

6:08We really need to build that foundational intuition

6:11about how the models work.

6:13So let me write ML here,

6:15and then here it's going to be AI engineering,

6:18and we're going to focus more on deployment,

6:20monitoring, and system behavior.

6:24Partly because AI engineers are not preparing the data,

6:27modeling the data, training it, and evaluating it.

6:31That's already been done with deep learning

6:33to create that foundation model

6:35so that the AI engineers can then focus on deploying a model,

6:39normally a foundation model,

6:41with monitoring and system behavior

6:43using system prompts and RAG and all kinds of techniques

6:47that we'll get into later in this course.

6:49So don't feel as if you have to master

6:52this whole pipeline right now.

6:54Instead, we're going to zoom in on different parts

6:57at the appropriate times in the course

6:59before we get into frameworks like PyTorch and TensorFlow.

7:04All right, so that's it for this video.

7:05And in the next video,

7:06we're going to look at machine learning engineers

7:09and ask a question,

7:10how do we make an AI system useful,

7:13reliable, and usable in the real world?

7:16All right, I'll see you in that next video.

0:00Machine learning and AI engineering overlapped, but a machine learning engineer might ask,

0:05how do we get a model to learn patterns from data? I would say an aspiring machine learning engineer

0:10and an aspiring AI engineer might ask, how do we make an AI system that's actually useful,

0:16reliable, and usable in a real world product? Because that's what everybody's really focusing

0:22in on right now, much more than machine learning. And if you're asking yourself,

0:27well, which is the best of both worlds, right? Ideally, if you have a data scientist that also

0:34performs AI engineering, that is the best of both worlds because a data scientist really focuses on

0:41cleaning data, preparing data, modeling that data in different ways, which includes machine learning,

0:47but there are many other techniques. So this is really the sweet spot, but it might be overkill.

0:52For some applications, all you need is AI engineering. And for some applications,

0:57it's really focused on model building. And these differences might sound small at first,

1:01but it really does affect almost everything. So I said they overlap. So let's go ahead and

1:06draw that here. So here we have machine learning and here we have AI engineering.

1:12So let's take a look at some examples. So let's say I gathered a bunch of labeled emails,

1:17and then I want to train a model. This is almost like pseudocode. And the goal here is to classify.

1:23We want to classify them as, let's say, binary classifier, spam, not spam, because this makes

1:29sense. If it's spam, it goes into the spam folder. If it's not spam, well, you can read it. Nothing

1:35happens to that email. It doesn't get rerouted. And I would say that this is a classic machine

1:40learning problem. And you might ask yourself, well, what is a machine learning problem and

1:45what is not a machine learning problem? This is a good machine learning problem because we can learn

1:51from the data and then start to make predictions on new and unseen emails, right? Data. However,

1:58music is not a problem that needs to be solved. And there are a lot of AI engineers and machine

2:04learning engineers that are trying to solve music, but it's not a problem. You can do creative things

2:10with it. For example, musicians don't like to play music, right? Like boring elevator music.

2:16Well, that's a great job for a machine learning engineer to create a generative AI, again, AI

2:22engineering with fine tuning on lots of audio data to create corporate music or elevator music. But

2:30the act of creativity is a human activity. So when you start replacing human activities that bring

2:37joy and value to society with a robot, a machine, that's not a good problem. Right now it's happening

2:44a lot. However, spam and not spam, that is a great job for a machine learning model and not a human

2:50being to look at that. So some problems are problems that need to be solved and some are not.

2:56But I did want to mention generative AI is very powerful, but we really need to think about the

3:01problems that we're solving. And now let's think about the pipeline here. Going back to this

3:06classic machine learning problem, it's already been solved, but we're using this as an example.

3:10When we're considering the pipeline, you want to care about the quality of the dataset,

3:16the features or the representations, the model that you're going to use, and the training process.

3:21And finally, the most important is the evaluation metrics, because you don't want to say, oh, my

3:27model is 100% accurate. That is very rare. When you do the evaluation, you might notice that it's

3:32something like 70%. And for spam, it goes down a little bit low. We want to get that higher, right?

3:37But in some cases, 70% will be acceptable. In some cases, you would need it to be 99.9% for surgery,

3:44right? So it really depends how performant your model needs to be. But now let's think about AI

3:49engineering. Here, let's build a workspace assistant, and this could be Google Gemini or Microsoft

3:57Copilot inside of the Microsoft suite of applications. Those already exist, but let's say

4:02that we're building a local environment assistant for our workspace. This could be useful for a

4:08startup, and you can have the workspace assistant read policy documents, answer employee questions.

4:15Let me put this here, internal. Also, citations, so we know that the model is actually referencing

4:22either policy or existing documentations, and we need guardrails. Those guardrails ensure that when

4:28the model doesn't act as you would expect, we can gracefully manage those. And that way, we can

4:34avoid confident misinformation, which is everywhere right now. But you definitely want to create

4:42guardrails and avoid that in an internal workspace assistant. So this is going to be much closer to

4:49AI engineering than machine learning. And now I care about the foundation model, not the model that

4:56I'm going to build. So here, with machine learning, it's model building. And here, it's really about

5:01the foundation model. Let's say you're interested in programming or you're interested in text.

5:07Depending on the model, you're going to get different behaviors, so you can test them out.

5:11You might choose Cloud Anthropic. You might choose Grok for programming. Those are foundation models.

5:18And then you want to think about prompt structure because you're not programming with code. You do

5:24that with machine learning, but with AI engineering, it's prompt engineering. You're programming in

5:30English or whatever plain language you speak. And we're also going to talk about retrieval

5:35augmented generation, and that's where you provide policy or internal documents that can be used

5:42for that model, this foundation model, to provide outputs that include proprietary information.

5:48Because models, foundation models, aren't trained on your company data, especially if it's a startup,

5:55right? That data doesn't really even exist. So how do you get that into the model? If you're

5:59a machine learning engineer, you could fine tune it with that, meaning take your internal documents

6:04and train it. But that is very expensive compared to AI engineering, and it's more of a permanent

6:10solution. Whereas AI engineering is great for internal docs because they change all the time.

6:16So all you have to do is upload whatever documents that you have, and that way,

6:20RAG is better than fine tuning. So we'll cover fine tuning and RAG later. We also

6:25care about fallback behavior. If something goes wrong, the output format is critical, right?

6:31You want the output to look good and to be useful, and you don't want it to take forever. You want

6:35to make sure that it's fast and you want to build trust. You want to make sure that you're handling

6:40things like PII or protected health information. All of that needs to be protected from these

6:46models because it is a liability to put information into a foundation model. For example,

6:53if you're using something like chat GPT, that model indefinitely stores a lot of data and it

7:00can leak that data and be used for training. The same thing goes for Grok, right? That's another

7:06model, but Grok is great for X and real-time information. Gemini from Google and Copilot

7:13with Microsoft, those are better for compliance because those models are built inside of the

7:20identity boundary for your organization. So there's a safe connection between those documents,

7:25not so much for Grok or for chat GPT. They're outside of your organization. So definitely

7:32consider the right foundation model in this scenario. And what else do we care about?

7:36System reliability. That would be monitoring. So what changed between these two? In the first case,

7:44machine learning, this is really about learning, right? So the model learns from the data. And in

7:50the second case for engineering, this is mostly about system behavior. You're guiding this

7:57foundation model to behave in the way that you want. But here's another way to think about it.

8:01Machine learning is a workflow that often ends with here's my trained model and here are the

8:07evaluation results, right? Let's say that it's 70% and that's good enough because you're just

8:13dealing with mechanical machines that do some sort of simple agriculture and it doesn't have to have

8:20a very high accuracy percentage. And in here, this is maybe you care about having it be very accurate.

8:27And that's more about machine learning, really looking at those evaluation metrics. Now,

8:32when we go to AI engineering, the workflow often ends with here's the working application. Here's

8:38how the model behaves inside of it. And here's how we tested reliability. And more importantly,

8:44here is how users are experiencing this moments where you can do thumbs up and thumbs down.

8:50That is called reinforcement learning with human feedback, RLHF. And you have to create system

8:56prompts to steer that behavior. But you also have to monitor the exchanges, the conversations to make

9:03sure that there is not inappropriate information, failures of security. You start very conservative.

9:10And as the performance improves in these AI engineered models, then you can rely on it more

9:15and more and increase the users because you've tested it. It's a very different approach. So not

9:22every AI system needs you to train the model from scratch and not every predictive problem

9:27needs a giant language model. And not every useful data product is AI engineered. Also,

9:33not every AI engineered workflow is machine learning heavy. There is a little bit of an

9:39overlap where you might fine tune a model and create an AI engineered product, or you create

9:46your own foundation model and you create your own AI engineered product, which is very expensive

9:53and is done by large companies with large budgets. All right, so now let's look at machine learning

9:59at a high level in the next video.

0:00In the last video, we looked at AI engineering and machine learning and how they overlap.

0:05And in this video, we're going to cover machine learning at a high level and AI

0:09engineering at a high level. So one way to define machine learning is to say that

0:14it's a way of building systems that learn patterns from examples instead of depending on

0:21handwritten rules. So it is not software 1.0, which is traditional programming.

0:28The models learn from the data, meaning that it's software 2.0. And again, that could be

0:32machine learning or deep learning. And this is an umbrella term machine learning because

0:37deep learning is a subset of machine learning. It's the newer version of machine learning.

0:42And it uses those neural networks and machine learning up here uses algorithms,

0:47but they both learn from the data. That's the core idea. So in traditional programming here,

0:52we're telling the computer exactly what to do. We write the instructions, we define the conditions

0:57and we control the branching of logic and machine learning or software 2.0. We still write the code,

1:04but the code is no longer the whole intelligent system. Instead, we use code to define the

1:09training process. So we're really just feeding the system some data and let it adjust the

1:14internal parameters in those hidden layers that we discussed. And then we can evaluate

1:19whether it learned enough to be useful, to generalize on new and unseen samples.

1:25So that's inference. So the heart of machine learning is not just predictions.

1:30It's learning from the data in a way that can extend beyond the exact examples seen during

1:36training. So you have training data and then you get a 75% accuracy rate. And then you measure that

1:43data, right? To see if it can make predictions. And that means that you're going to show it data

1:48that is not seen before. And then you're going to get an evaluation metric there. And let's say it's

1:5372%. Very close to this. That's pretty good. So you have a performant model for whatever the

2:01metric threshold is. And that is 75% in this case. Now let's talk more about machine learning

2:08prediction. And again, we're doing this in English because we're going to do some pseudocode very

2:12soon. Let's say you want to predict a number. So in that case, we could use machine learning

2:18for house price prediction. Also sales forecasting, temperature estimation, and maybe demand. So you

2:26want to predict and demand prediction. This is often called regression. But if I want to predict

2:33a category, let me go ahead and put an arrow here for number. And for category, we might use machine

2:40learning for spam, not spam. It can also be for fraud versus not fraud. And we can continue with

2:47this churn, not churn, positive or negative sentiment. I'm going to write this one because

2:53it's also called sentiment analysis. If you ever hear that you're just predicting, in this case,

2:58binary, positive or negative sentiment, or maybe cat versus dog pictures, or hot dog versus

3:05hamburger, or handwritten digits. This is often called classification. And you already know that

3:11machine learning is not one single algorithm. It's a family of approaches for learning useful

3:17patterns from data. And those patterns live inside of models. So let's say the model is simple,

3:24and you need to perform regression like we have here. It might be linear regression or logistic

3:29regression. But if you're dealing with more complex data, you might use neural networks.

3:35All right. So this is a pretty good definition of machine learning and the different kinds of

3:39predictions we can make with numbers and categories. Now, in your mental model, try to

3:45remember the pipeline. So the workflow usually involves the same ingredients, data, labels,

3:52or targets, a model, training, evaluation, and then prediction, right? Once you have that,

3:58you can start making predictions. That's also called inference. And at this point, you might

4:02be saying, okay, I get it conceptually, right? Classification and regression, numbers and

4:08categories. All right. House prices or sentiment analysis. But let's put this into concrete terms.

4:14And let me show you the numbers that we're going to predict for regression. Let's say that we're

4:18working with student data, and we're going to have only two inputs in this case. And that could

4:24be hours studied in the attendance. Let's call this rate like a percentage. So these are the two

4:31inputs. And now we want a target. And so that target is going to be if a student passes or

4:38fails. Very simple. And now thinking about the machine learning workflow, we might do something

4:43like this. So it's very common to use X as the variable for your inputs. I'm not going to get it

4:49too detailed into the code, but let's create a representation of some code here. And in the first

4:55one, you might have two and then 0.60. Okay. And then you might have another one and that says like

5:03five, 0.80. And you're going to have n number of samples. So right now we have two, but that's very

5:10little. You need a lot more than two. So this represents the hour study two and the attendance

5:16rate in this case is 60%. Now the target here is represented with lowercase y. So we'll talk about

5:24the reasons why we use uppercase X and lowercase X later, but for right now, X and Y input and output.

5:32So the output is also called the target. So let's say we have an output for these two. So the first

5:37one is going to be fail. And the second one is going to be pass. So let's define these one and

5:43zero. All right. So we have the input examples here with X and then the labels, the output or

5:49the target is why, but where's the intelligence? How is it learning? We haven't done that yet. So

5:55this is the setup for learning, right? And later we choose the model, then we train it. And then

6:01we see if it can predict the correct labels for these cases. So when I say machine learning, I

6:06want you to think data in pattern, learn prediction out, and that's it. That's a good mental model for

6:13now. Now switching gears to AI engineering at a high level, we can compare them. If machine learning

6:20is about getting a model to learn patterns from data, then AI engineering is really about useful

6:25systems built around AI behavior. So that's the AI engineering part. And that means it's less about

6:33inventing a new model from scratch and more about making an intelligent behavior that actually works

6:40inside a real application or a workflow or a product. So the kinds of questions that an AI

6:46engineer cares about are which foundation model is best. Let's say you want to use real-time data

6:54that comes from all over the world. Well, Grok is a good idea because it's connected to X that has a

7:00lot of real-time data. The problem there is that there's a lot of noise. So you have to find a way

7:06to amplify the signal. But let's say you want to use a model, foundation model, as a writing

7:13assistant. A lot of people say that Claude from Anthropic is a good way to go, is good at writing.

7:19So you pick that foundation model. And then you say, do we need a classifier or a language model

7:24or a modal model? Remember that one? You can give it multiple types of inputs like audio, image,

7:31and then get some kind of video that goes out. Another question might be, how do we structure

7:36the prompts? So that is prompt engineering, which includes something called system prompts,

7:43which is also part of prompt engineering. This is how we steer the model. If you're using a product

7:49like Whoop, I use this and I can show it to you right now. So right here, you can see this band,

7:55right? It's just a band. There's nothing to read. It's just tracking my health data. And on this

8:00side, I have a watch. I really turn off most of the features on the watch. And I just use it to

8:05say, I'm going to start this kind of workout. And then this one over here will record that data.

8:11In that case, the prompt engineering and system prompts, you might say something like you are an

8:15expert trainer for athletes. And you might use a system prompt that says you are a health

8:22application trainer and you provide concise, helpful information based on user exercise data.

8:29And then that way you get an LLM or an assistant that is highly effective. And Whoop is one of the

8:35best AI models that I've used because it makes sense. It cleans the data and provides really

8:41actionable, concise information that I can use on a daily basis. Everybody can be a data scientist

8:48with this application. For me, I can talk to the model and we can actually perform experiments

8:53conversationally. I can do an A-B test. These are things that we do as data analysts and data

8:58scientists, a little bit outside of machine learning and AI engineering. But again, this is

9:03what I was saying. These models meet you where you're at. So I'm able to use it as a data scientist,

9:09but a casual user can say, help me use data science techniques to improve my fitness.

9:15And that's sort of what it's doing in the system prompt behind the scenes. And it's very effective.

9:20And that is also all about context, which Whoop is recording on my wrist. How do we give the model

9:26the right context for Whoop's case? Well, they give me a band and that's how it's done. And an

9:32AI engineer also might care about and ask, how do we make the output predictable enough for real

9:38users? And how do we reduce hallucinations or low quality responses? Misinformation, prompt

9:45engineering is a big part of that. You might ask how to evaluate whether the system is actually

9:50useful. In that case, there are frameworks like Lang chain that allow you to monitor the

9:56interactions and make any adjustments as you go. Because these models are not measuring the

10:03performance from training data to evaluation data, you're training the data or you're improving the

10:09AI engineered model by looking at the chat conversation. And finally, how do we integrate

10:15this model into an app, meaning the foundation model? Maybe it's going to be on a website or

10:21it's an assistant or an internal tool. All right. So we know that machine learning, the focus is on

10:27training data labels, maybe the model improvements, the evaluation metrics and inference or predictions.

10:34However, in AI engineering, the focus is more on foundation models or model orchestration,

10:42meaning you can use more than one foundation model. Also, user interaction. We want to make

10:46sure that the model isn't saying inappropriate things and hallucinating and giving the user

10:53wrong information. And that ties into reliability. Cost is a very big one. These tokens can get very

11:01expensive. So you really need to monitor the cost as well. And speed is latency. This is a very

11:07important consideration, which ultimately leads to system behavior, which we can steer with prompt

11:13engineering. And in the previous machine learning workflow, we built a spam binary classifier. So

11:20spam, not spam. Right. So that was logistic regression in this case. But if I build an AI

11:27internal assistant that can search internal documents, answer questions, cite the resources,

11:34follow tone instructions. Again, that is prompt engineering and return a nice output for a positive

11:40user experience. This is really AI engineering. Right. Also, we can create a chat bot that greets

11:46users and pulls account details from a tool and answers simple questions and then escalates harder

11:53cases safely. That's also AI engineering. So very similar to the Whoop app. Right. I've actually said,

11:59hey, send us honest feedback. And it says, I'll send that for you and make sure that we use this

12:04information to improve. There's many ways of doing this. You can also build a writing assistant that

12:10rewrites rough internal notes into polished emails and follows the not only the company tone,

12:16but compliance. Right. That is also a good example of AI engineering. So AI engineering is not just

12:23using AI. It's designing systems that make AI useful. A big difference. And it's way more than

12:31chatbots. People think that, you know, these chatbots are AI and that's what's going to turn

12:36into AGI. You may have heard of AGI. That is artificial general intelligence. It's the next

12:42level. But chatbots are just one sliver of generative AI and there's machine learning and

12:48deep learning and so on. All right. So that's it for this video. And in the next video, we're going

12:52to talk about data science, data analytics, and also data cleaning, which is super important.

12:58See you there.

0:00Welcome back. Now it's time to widen the fields a little bit, because if we only talk about

0:05machine learning and AI engineering, we can accidentally make it sound like it's a whole

0:10field that starts with models and it doesn't. A lot of the real value happens earlier. And this

0:17is when people are still trying to understand the data, define the problem and figure out whether

0:22a model is even necessary. And this is precisely where data analysis and data science comes in.

0:29So let's check them out one by one. Data analysis is more about exploring data,

0:34finding patterns manually or with tools, answering questions and ultimately helping people understand

0:41what is actually happening. So they might summarize trends, compare groups. Again,

0:47this is very common in business intelligence or BI. They might check distributions. For example,

0:53you might get a normal distribution with most of the values here in the middle where everybody is

0:58getting a passing score of, let's say, 60 and very few are getting 100 and very few are getting zero.

1:05Again, these are ways of finding patterns. They might also investigate anomalies and even building

1:11dashboards. That's a very common part of data analysis. So what patterns are they discovering?

1:17So it could be churn is rising in one customer segment or that support response times are

1:23getting worse on Mondays. Makes sense. And that sales performance is going to vary by region or

1:29the key metric is being calculated incorrectly. That is incredibly valuable work. And not every

1:35one of those insights needs a machine learning model. In fact, you could do this without it

1:40because historically data analysts have not been using machine learning models until recently.

1:46We can say conversely, data science. This is a much broader field. So we can say data science

1:53is like this and you might have data analytics over here. You might have machine learning over

1:58here and you might have AI engineering over here. So data science touches a lot of these domains.

2:05So depending on the team that might include all of data analysis plus experiments, let's perform

2:13an A-B test to see if one website page works better than another or maybe a button to increase sales.

2:21Also, it is statistics heavy and that is for statistical reasoning. Another way to describe

2:29machine learning is statistical learning because here this is done historically without machine

2:34learning models, but now you can use machine learning for statistical learning where they

2:39learn the patterns on their own. Also feature design, maybe combining different columns to

2:45create new features, but there's also modeling. Not all data scientists intersect with machine

2:52learning to do modeling, but I do. And I started my journey as a full stack developer working

2:58through data analysis and then working through data science and then machine learning and then

3:03deep learning and then AI engineering. So you can think of that as another full stack. So there's

3:09full stack engineering and now there's full stack AI, which I believe includes full stack,

3:14data science, machine learning, and AI engineering. That's because I have ambitious goals and I'm an

3:21educator, so I've gone sequentially through that path, but that is not the most common path.

3:26Different jobs require you to do different things and different backgrounds will lend themselves to

3:31different paths. So maybe data analysis is all you needed for your role or maybe machine learning

3:37engineering is all you needed because there was a data science team and they would give you these

3:41clean data sets and then you would build models on them. You didn't really have to do data analysis or

3:47feature decisions and things like that. Think of it like this, data work or data practitioners use

3:53layers. Sometimes the right move is analysis. Sometimes it's modeling. Sometimes it's an

3:58experiment or maybe improving labels or a machine learning model. And sometimes it becomes AI

4:05engineering. So let's take an example. Let's say a company says, we need AI to reduce customer

4:12churn. All right. So when I see that, I would ask, why do you need AI to reduce churn? I would

4:18prefer a question. We need to reduce customer churn and not add AI, but let's say they did.

4:25And there's a reason for it. And I signed off on it. So what are the questions that we might ask?

4:30So a data analyst or data scientist might ask, what exactly counts as churn? So the data

4:36definition is the target label consistent. So we're thinking about the labels, right? Because

4:42we're going to train a model and are we measuring the right time window? So the data window, because

4:48maybe we're using historical data and there's a lot of data, but we just want to focus on this window

4:54because what happened there? So is the churn problem actually concentrated in a few specific

5:00customer types or do we need a model or maybe first we need a better dashboard and a cleaner

5:06definition. This can bring incredible value to your team because it reduces a lot of the problems

5:11that happens before it gets to a machine learning engineer. And before it gets to an AI engineer,

5:17garbage in garbage out, this first step is critical. And maybe you're responsible for

5:23all of this on your team. It really depends. Maybe you're only building models or maybe you're

5:28only doing AI engineering. Just wanted to put that out there to really clarify this context.

5:34So whether or not you have a data science team, we have to think about data and analytics and

5:38data science and make sure we're covering all the bases and short data analysis and data science

5:43provide foundations that makes the models or the products smarter and more efficient and performant.

5:50And now we're going to segue to data cleaning because it's one of the least glamorous,

5:55but most important parts of the whole stack data cleaning. I have to say, it seems like, wow,

6:02it's such an unglamorous thing, like being a data janitor. I can't tell you how many times I've

6:07heard that from data scientists. And it's true. It feels like we're cleaning the different sort

6:12of environments for the data in the data itself. But if you don't have that background cleaning

6:17data, let's talk about that real quick. And what we see out there, especially with vibe coding,

6:21is that people want to build models. They want to use different frameworks and they want to

6:27add some prompts and they want a really cool app or a demo. But if the data is dirty,

6:33the whole thing crumbles. And that's not only for vibe coding, but for companies and organizations

6:40and even enterprise. You'd be surprised how really big companies are skipping on these

6:46elementary steps, creating security vulnerabilities and poor implementation of AI. Here's why the

6:54model learns from data that you give it, not the data that you wish you had. So what does this mean?

6:59Data cleaning. When you get raw data, I can say with a high degree of certainty that it's almost

7:05never going to be ready for modeling. So what does that mean? The kinds of problems that we're

7:10talking about here are things like missing values, duplicates, often rows, inconsistent values,

7:18inconsistent labels, mixed data types. A very common problem here are mixed date formats. You're

7:24sending it as a string meaning text, but it needs to be an actual date time object. When you don't

7:31do that, it's going to break stuff. Also, you can have broken text encoding, noisy text, mislabeled

7:38examples. Remember hot dog and hamburger? How about if we said hot dog and beef stroganoff?

7:44Well, one of those examples called beef stroganoff is going to mess with the performance, right?

7:50And unlabeled classes. We had 100 images of hamburger and 100 of hot dog. But if you had

7:561000 of hot dog, guess who's going to win all the time? Almost everything is going to be a hot dog

8:00to this model, right? So you want to make sure that you have balanced classes. And there's also

8:05data leakage between training and testing data, a lot to take into consideration. For example,

8:12the inconsistent labels, let's say you have spam, all lowercase, or spam, all uppercase or spam,

8:19title case, and then another one just says junk altogether. Well, all of these need to be

8:25standardized, somebody needs to merge the data sets. And now those categories mean slightly

8:30different things. So we need to fix these very small details, including duplicates,

8:35maybe the same email appears many times because there was some sort of export problem, or the

8:41training data set accidentally included a version of data that leaks the answer. These are all

8:47very important considerations. Alright, so that's it for this video, I just wanted to make sure we

8:51covered this because it's happening right now. It's a real serious problem. But if you address

8:57it and you have these skills, you're going to bring a lot of value to your organization.

9:01I'll see you in the next video on the pseudocode crash course.

0:00Welcome to this video. Before you move into the final challenge, I want to give you a quick but

0:05very important tool, and that is pseudocode. Pseudocode means writing the logic of a system

0:11in plain English before you write the real code. You can do this with software 1.0, rules by hand,

0:18software 2.0, which is the learning of data from either a machine learning model or a deep learning

0:25model, and even software 3.0 when you're doing AI engineering or working with generative AI.

0:31If you're a vibe coder, all of this can be very helpful. And this is where I differ from a lot

0:36of the online courses on machine learning. They really focus on getting you to write code

0:41immediately using frameworks even and working with the hand digit classifier or different popular

0:48data sets for machine learning. You can either have the code written for you or you can write

0:53out the code yourself and then you execute their code and they explain it as they go, right? And

0:59then you get some sort of a model and some kind of an accuracy. It's like, oh, wow, 85% accuracy,

1:05but so much has been abstracted. So my approach is to leverage pseudocode and talk about it so

1:11that you're understanding the logic of what's happening before you get into the syntax,

1:16because that syntax will also change from framework to framework. And it's easier to

1:22think about one thing instead of two things. That's just how the human brain is wired. So if

1:27you have a good understanding of the pseudocode or the logic, then you can just extend that to

1:32the different frameworks and you can be syntax agnostic at that point. That is going to be very,

1:38very helpful when the performance is not 85% because someone wrote it for you, even a model

1:44or an online platform. It's up to you to understand why it's 85%. And usually it's

1:50not going to be 85%. It might be 50%, just as good as guessing. And you want to understand the

1:56logic so that you can increase the performance because you understand what is happening in those

2:00hidden layers where you have activation functions and all kinds of hyper parameters, like the

2:06learning rate, which are very key to the performance of these models. So I like to do this in plain

2:12language. First, the goal here is to give you a deep system understanding so that you can express

2:18yourself in Python or any other language and definitely treat this as real technical thinking.

2:24This is not a fake version of coding, right? This is as good as it gets. So here's why it's helpful.

2:29When you start building a model, you might ask, what is the input, right? So you slow down and

2:35think about the system for a minute. And that's exactly what you want to do. That's a great way

2:39to start. Also, what happens to the input, what decisions are being made and what should the

2:45output look like, right? And how do you know if the model or system works? So let's think about

2:52the pass fail system for student scores that we talked about earlier. In that example, you can

2:57write, read the student score. And I'm going to assume that you've cleaned the data. All right.

3:03And now you're ready to ask these questions. And if the score is greater dot, dot, dot, and this

3:09could be any logic, it could be is greater than or equal to 70 return pass. I'm just going to

3:15leave these dots for some kind of logic and then otherwise equals fail. Very simple. So this is not

3:22executable in Python, but it's absolutely clear syntax logic. If you put this into a chat bot,

3:27you would probably get code that works. It's not terribly difficult, especially if you provide the

3:32student scores. And if you're thinking about general machine learning, you might say

3:37gathered label examples, then split the data into training and test sets, choose a model,

3:42train the model on the training data, evaluate the model on unseen test data, improve the data

3:49or the model if the results are weak. So ideally you would not have to go back and clean the data

3:56because your model is not performing well, that should be done. However, if the data is clean

4:01and you have solid data and the model is not performing well, then you know where to look.

4:06And for general AI engineering systems, pseudocode might sound like receive the user's

4:11request, which would be a prompt, gather relevant context. This might be retrieval,

4:16augmented generation, send the request in context to a model, return the output in a useful format,

4:23and then log the result for review or evaluation. So very similar, but it's completely different

4:30in terms of what the tasks are. So the questions mirrors the workflow, and that's the takeaway.

4:37Pseudocode lets you think at the right level before implementation starts because it can

4:43easily distract you. So in the challenge, we're going to try this out and I'm going to

4:50give you a colab notebook that has completed code and you have to write the pseudocode above,

4:55and it can be as simple or as complex as you like. Just make sure that it represents what's

5:01happening in the code below. So it's kind of like reverse, right? You write the pseudocode,

5:06that's your challenge, and the code already exists, but it has to sync up with your logic.

5:10And I think that's going to be a great way to make some connections and it should be a lot of fun.

5:14So I'll see you in that next video, which is the challenge.

0:00Congrats and welcome to the final video and the challenge.

0:03Your task is to solve the same problem three different ways

0:07so that you can see very clearly how the paradigm changes

0:11the logic, the workflow and the strength of this solution.

0:16So here is the shared problem.

0:18Classify an email as spam or not spam.

0:22All right. And that's it.

0:23This is a great challenge because it's familiar, it's realistic,

0:28and it's simple enough to compare across different software styles

0:32without turning it into a giant project.

0:34So we're going to take that one task and use software

0:391.0, 2.0 and 3.0.

0:43In the previous skill, the challenge was really focused on this.

0:47So we know that software 1.0 is where you write the rules yourself.

0:51That is traditional programming.

0:53Software 2.0 is where the system or the model learns

0:56from labeled examples, which includes machine learning or deep learning.

1:01And 3.0, you steer a pre-trained foundation model

1:05or model with plain language.

1:08This could be generative AI or all the way to AI engineering.

1:13Generative AI, just to clarify, is when you talk to a chat bot

1:16or some generative AI application and you use a prompt

1:21and then you get an output, right?

1:23Give me a picture of a rocket that is not made out of metal,

1:27but instead it's a banana flying through the sky.

1:30You can have fun with that.

1:31And that's basically gen AI.

1:34And as you know, now, AI engineering is a lot more.

1:37This is where you steer the model using many, many techniques,

1:40including programming, but also guardrails and many other things.

1:45Just wanted to make that general distinction that this very simple part

1:50really is using gen AI.

1:52But here, gen AI also includes AI engineering.

1:56And what's great about this is that the problem here stays the same,

1:59but the mechanics change.

2:02So let me walk you through exactly how to complete this challenge using pseudocode.

2:06So this notebook is going to be made available to you.

2:09All you have to do is click on that link.

2:11And when this opens up, you want to go to file and save a copy and drive.

2:15You might be prompted.

2:16Let me show you what I mean.

2:18Here, I opened this same notebook in a private tab

2:22because now you see changes will not be saved.

2:26I'm not signed in.

2:27And it says copy to drive.

2:29So you don't have to go here, but you could do that to go save copy and drive.

2:33But you can just as easy just click on copy to drive.

2:36When you do this, you have to sign in,

2:38but then you can actually choose where you want it to go.

2:40OK, so I'm going to close this

2:42and then we get back to our main notebook where I am logged in.

2:45And so let's go through this.

2:47This first one right here is rule based spam classifier.

2:51And here is a section for pseudocode.

2:53The solution code, meaning one through five written solutions

2:57are provided here for your benefit.

3:00And you can check those after you complete the first five.

3:03So this is a self graded collab notebook challenge.

3:07And here is the text out with the solution.

3:09So go ahead and click on this carrot to expand it.

3:12Once you've filled that one through five.

3:14So what do I mean by one through five?

3:16So here in this first one, this is really what the system is going to receive.

3:21And this next one is what the system does.

3:24And then this could focus on the output.

3:27And here you can say, well, here's the strength of the system.

3:31And here is a weakness.

3:33This is my approach, but you can do this however you like.

3:36You can add more numbers or you can reduce the number of numbers

3:39if you don't need that many pseudocode steps.

3:42That structure is important because it forces you to think like a systems designer.

3:47For example, in software 1.0, your pseudocode is going to include

3:51something about email text, right, all the way to the output of spam or not spam.

3:57So that's pretty straightforward.

3:59Now, let me take you through each code example.

4:01So here's software 1.0 rule based spam classifier.

4:04And here we have a simple function.

4:06And what is important to understand here is that we're giving the system

4:10in this case, very simple phrases.

4:13This is a focused educational toy data set, right?

4:17We just have five very simple things.

4:19But imagine spam phrases could be very large.

4:23So it's good for deep learning, in my opinion, or machine learning,

4:26depending on how many emails you have and what the objectives are.

4:30So here we're going to make sure that the email is lowercase.

4:33And we're going to look at the suspicious phrases.

4:36And if that phrase is found as lowercase right here and email lower,

4:40then return spam, because if this is in any of your emails, pretty sure that it's spam.

4:45Otherwise, it's not spam.

4:46So here are the emails that it's going to check.

4:49So it's checking to see if any of these have these suspicious phrases.

4:53And then this is the output.

4:55I'm not going to do the output. I'll let you do that.

4:57And for software 2.0, I'm not going to go through the pseudocode.

5:00I'll leave it up to you.

5:01This time I added six steps because that's what we're using in the output.

5:05And I want you to see that this is learned behavior.

5:08Instead of writing the rule manually, the model learns from examples from data.

5:12Here, we're going to use scikit-learn.

5:14All right. So we're not using a framework just yet.

5:16And we're not going to get into scikit-learn because this is an introduction

5:20to machine learning.

5:21However, if you want to get into machine learning

5:24and you want a proper machine learning course, please let us know in the comments.

5:28This is going to help us to understand what you are looking for

5:32as far as the courses in the context of machine learning.

5:35And if you want a course all on AI engineering, again, let us know

5:39and we'd be happy to take a look at that.

5:41And if there's enough interest, we will make that course for you.

5:44All right. So we have a similar system here.

5:46We have emails.

5:47So this is our small labeled data set.

5:49And here are the labels spam, spam, not spam, not spam, spam, not spam.

5:54And here we're going to build a simple text classifier.

5:57Here's where we train the model.

5:59So model that fit and we're using the emails and the labels. Right.

6:02So the input and the output here, we're going to classify these new emails.

6:06So new emails, action required.

6:08Confirm your direct deposit information.

6:11And here is another email.

6:13And then the predictions.

6:14And here is a pseudocode solution.

6:16And finally, for software 3.0, you're going to do the same thing

6:20with the pseudocode and then AI engineering.

6:22So what we're doing here with 3.0 is steering the model behavior.

6:26We're not retraining the model.

6:27We're just prompting. Right.

6:29That's how we steer with plain language.

6:31And down here we have the solution for the pseudocode.

6:34Be sure to look at that after you fill these five out.

6:37Again, you can make these as short or as long as you need to.

6:39And then finally, we have the bonus.

6:41So API style example.

6:43The reason I'm adding this here is it's very common

6:46to see something like this in AI engineering. Right.

6:49Here is a structured prompt.

6:51So this is a system prompt.

6:52You are classifying workplace emails.

6:54Now it knows how to behave and you're going to return.

6:57This is like focusing on the output label spam or not.

7:01Spam reason one short sentence.

7:04And this is where the generative AI, the foundation model can actually do that

7:08because it's being instructed very clearly.

7:11And then finally, you get the email and the email text,

7:13and then you would do some sort of response talking to that API,

7:17just including this here for your reference.

7:19OK, the exact SDK will vary by provider, but the pattern is pretty much the same.

7:24We close this and that's it for this skill.

7:27So until next time, I hope this has been informative.

7:30And I'd like to thank you for viewing.