AI Engineer Roadmap – How to Learn AI in 2025

this AI engineering road map takes you from core fundamentals to Advanced AI implementations it covers essential mathematics machine learning deep

learning and large language models providing you with the exact skills needed to thrive as an AI engineer in 2025 whether you're starting fresh or

upgrading your skills this road map offers a clear path to success with hands-on experience and Industry relevant insights T from lunar Tech

developed this course imagine being at the Forefront of one of the most transformative fields of our time where technology meets

Innovation and changes the world welcome to the AI engineering road map of 2025 my name is D Vasan from lunar Tech

and I'm absolutely exciting to be here with you today to dive into this highly requested topic together we will will explore everything that you need to know

to navigate this exciting world of artificial intelligence and AI engineering to set yourself up for success in this field in this video we are going to break down the step-by-step

road map for becoming a worldclass AI engineer here is what we are going to cover first we will Define what AI engineering is and how it feds into this

broader ecosystem of AI and data science next we will explore the real world applications of AI engineering showcasing its really strong power

transformative impact across different Industries then we will dive into the must have versus nice to have skills helping you to understand exactly where

to focus your efforts and your time finally we will go to step-by-step process so the skill sets that you need to master outlining the essential topics

to help you become a job ready AI engineer this session is packed with unique insights and practical tips that

you won't find any URS so stay tuned without further Ado let's get started so let's start with the basics what is AI engineering AI engineering is

this practice of Designing building and deploying AI systems that solve real world problems it sits in this intersection of software engineering

machine learning and data science and here is how it fits into this broader Tech world and the ecosystem so the data scientists often focus on analyzing data

or predicting something or developing models AI Engineers take these models and make them work in the real world settings and with much more advanced

models they create systems that process data make decisions and deliver actionable insights for example in the healthcare a data scientist might develop a machine learning model to

detect the tumors in x-rays an AI engineer brings this to the next level he ensures that the model is integrated into Hospital Systems runs in real time

and works reliably under different conditions also AI Engineers they work with much more advanced models like deep learning models or neural network based

models so data science principles system design optimization machine learning deep learning is what all combines into one place which is AI engineering it's

not just about building models it's about making sure that those models actually solve problems and deliver value for the business or this public

Enterprise and that's why AI engineering is such a critical role in today's Tech ecosystem it's where this Cutting Edge research meets the Practical industry

impactful implementation so bridging this gap between the research and the actual engineering so um AI engineering isn't

just limited to one field it's changing Industries all over the world let's look actually at some of the examples how AI

engineering is making an impact first up is the healthcare so AI systems are used to analyze medical images predict patients outcomes and also assist the

doctors in the drug Discovery or the patient care AI engineers build the systems to ensure that those are scalable reliable and efficient for real

world use next up is the finance from fraud detection to aloric trading AI processes

massive amount of financial data in real time engineers in this field they focus on creating secure efficient and realtime systems that can handle this

sensitive information real time like FR detection in the retail and e-commerce in the platforms like Amazon they use AI to personalize recommendations optimize

pricing and manage inventory AI Engineers they design algorithms and systems that drive this experiences next up is the entertainment of course the

streaming platforms like Netflix they rely on AI for personalized content recommendations jna tools like Dolly and chatbot chbt are changing now how the

creators produce content next up is the autonomous vehicles so self-driving cars they depend on AI for navigation object

detection and decision making AI Engineers they are the ones who design this algorithms and Hardware integration to make this autonomous Vehicle Systems

safe and reliable so these examples are just few of them and they show how different and impactful AI engineering is so whether you are passionate about

health care Finance Tech defense or any other creative industry there is a place for you in this field and that is actually why the AI engineering is so popular this day and it's going to be

one of the most independent Professionals in the next decade there are many Industries and companies who are currently Hing when it comes to the

salaries for AI Engineers those are highly competitive just 40 ENT roll they start around 80 up to

120k at least for the midlevel engineers this is uh 120k to 180k in us and where senior roles this can take all the way

from 200 up to 750k in the US dollar so let's now get into the actual skill set that you must know in order to become an AI engineer

and here I'm talking about becoming a worldclass well-rounded real AI engineer not just someone who does promp engineering real AI engineer not just

someone who does promp engineering and without knowing these different models uh just uses them but actually becomes someone who will create new algorithms who will create their own unicorns or

will become an AI and without knowing these different models uh just uses them but actually become someone who will create new algorithms who will create their own unicorns or will become an AI

engineer that works at this uh large Cutting Edge companies like open AI Tesla meta and many other Cutting Edge startups so first up is of course the

mathematics mathematics is a Fiel when it comes to traditional machine learning all the way to the most Cutting Edge AI that you see nowadays so um when it comes to mathematics there are different

topics from this field that you must know not the entire universe of mathematics or the super advanced stuff but really the fundamentals and um these

are selected topics from different uh levels so you cannot just say first level of University or second level of University of that specific study no it's a combination of these different

levels from this different fields and studies that you need to combine in one place learn it such that you can move on on to the next page and today I will

tell you which are those in a more detail such that you are left with a specific topics for you in mind to learn mathematics if you decided to do a

self-study and become an self faced AI engineer on your own so first up is the high school mathematics in here um you

can understand doing basic divisions how to solve an equation with uh squared unknowns so for example a square plus something you are able to uh calculate

the discriminant to find the solutions to that equation you know this different um geometric um terms like what is sinus

what is cosine what is tangent what is cotangent uh the Pythagorean theorem um basically all the topics from the high

school all the way to the last level next up is the uh linear algebra of course linear Al ra comes usually from the second uh year of econometric study

or applied mathematical and statistical studies and this field is really important for understanding not just the traditional machine learning but also the Deep learning which is really

important and it's a more advanced type of ml that powers today's most cutting gge applications including the GPT models the Transformers Etc so if you

want to know and understand the cycle of n networks the training how it's being optimized and how this entire neural networks structure works then you must

understand linear algebra so when it comes to linear algebra let me tell you specifically what I mean not the entire linear algebra but really to understand the norm of a vector this understanding

of vector and matrices the cartisian coordinate system that comes from um the high school but then here is also very relevant to understand where the vector

are how you can position the vectors in the cian coordinate system understand this idea of Norm versus alal and distance the uh Pythagorean theorem here again the

orthogonality um you also need to understand the vectors and operations so foundations of the vector the special vectors unit vectors um and also uh the

idea of dot product the application of the dot product the C squares equation also you need to understand the matrices and the solving of the linear systems

using this idea of matrices so here you need to have the foundations of linear systems and matrices you need to uh be able to add matrices multiply them to

compute a DOT product between matrices or between Matrix and a vector um also understanding of ging reduction the

reduced ulum form the row reduced ulum form the no space the c space the rank the full rank this all will be foundation for you to understand how

this their networks work um if you truly want to understand um the different deep learning and AI models you also need to have a good basis when it comes to

linear transformation and matrices so this algebraic lows for matrices uh including how um it actually works how you can uh solve a system with the

linear equations multiple of them using these different Transformations so what is for example the transpose of a matrix what is the inverse of a metrix and

apply these different uh rows and the rules from linear algebra uh also what is the determinant how you can calculate it what are the properties of determinant the transpose of matrices I

believe I just mentioned and then you also need to understand some topics from Advanced linear algebra like uh the

projections of vectors um the gr Schmid process the infamous process that you um need to understand uh the metrix factorization really important not just

for the Deep learning but also for the traditional machine learning or the things like metrix uh factorization that is used in the recommender systems so uh this part is

also very important to understand the QR de composition ion values igon vectors uh which is really important for understanding the principal comp quasis and dimensionality reduction also the

igon de composition which is based on igon values and igon vectors and understand the singular value the composition or the SVD which is really important part as part of traditional

machine learning so um this is what uh you need to know when it comes to the linear algebra and if you are looking for that

one place to learn linear algebra then uh last year uh we have published an entire 26 plus hour course that covers all these topics in one place it was

quite a popular course uh and highly demanded one and you can get also a certification once you completed so check out this course the fundamental s

linear algebra uh at the lunch. to also

uh go through all these topics uh follow it study it practice it and then get also a certification next up when it comes to

mathematics Beyond um the linear algebra and the um High School mathematics you also need to understand calculus this one is really important as well uh you

will need to have an understanding what are the gradients what are the derivatives how you can calculate derivatives how you can calculate the integrals not just with one n but with

two variables basically so double integrals um how you can uh use this uh derivatives and integrals when comes to

optimization this uh concept of the slope and uh optimization of the models using the gradients first order gradient and second order gradient in the context

of it how you can adjust the parameters for better accuracy and um just a traditional

calculus one and some calculus 2 so um this is um no-brainer when it comes to AI not just for advanced AI but for the traditional machine learning learning

for understanding these different models you must know calculus next up is the game theory not the entire universe of Game Theory not all the topics but there

are some topics from Game Theory which usually comes from third year of econometrical or ply mathematical studies is something that you must know

think about NES equilibrium or the mean Max strategy or this um um this game where um competing is actually resulting

in worse outcome than collaborating so uh this idea of NES equilibrium is really important for understanding one of the foundational generative AI models which is the

generative adversarial networks so for understanding one of this Genna models you will need to also have this uh couple of topics from game theory in

place all right so that's about the mathematics um and here I'm also not mentioning this foundational geometry topics which is usually also covered as

part of high school so once again the sign cosine the tangent how to work with with the different um angles the 90° angle what are these different values

for different angles and this common notation with the pi so what the pi represents the radians Etc once you comfortable with this mathematical topics the next topic that I would

suggest you to study is the statistics statistics is very important when it com comes to becoming a well-rounded AI professional to understand the um idea

of predicting the next word but all the way to the very basic machine learning uh having this basics of Statistics will be very helpful to you so here is the

list of topics that I would suggest you to study when it comes to statistics so first up of course understanding this concept of probabilities to know what

the probabilities are what is its concept uh why it is used for this concept of probability distribution functions the PDFs the cumulative

distribution functions or the cdfs and also um to understand uh what is this idea of sample why we use sample um versus

population um this idea of having a representative sample work with the data so understanding for example what are the random variables what is this idea

of experiment uh what are the probabilities um the uh criteria and qualities of probabilities what is the PDF or the probability distribution

function uh what is the cumulative uh distribution function this uh basic statistics like the mean the median the variance the standard deviation the mode

um and also how they can be calculated this um idea of covariance and correlation what is the difference between correlation and cation uh

understanding um how these different statistics can be used to describe your data and to tell a story about your data

and um also this idea of Sample versus population why we use sample um and why we um are unable for example to deal with a

population um and how this becomes relevant when it comes to this entire universe of data science um also understanding the bias theorem the

different rules when it comes to the probabilities like the conditional probability the idea of Independence between different random variables um

then I get into some Bic probability distribution functions especially the normal distribution function the baroli distribution function this idea of boli Trials the binomial distribution

function what is this connection between bomal distribution function and the binomial distribution function how it is used in these different concepts like

tossing a coin so basic statistics basically uh also understand uh the idea of uh linear regression and ordinary Le

squares what are these different uh conditions and assumptions that this ordinary squares is making when calculating and optimizing these

different um parameter estimates this idea of estimation versus um the unknown parameter the idea of error terms the error terms versus

residuals um and also this concept of gas Mark of theorem how it is used um and this comes usually from econometrics and the idea of parameters what are the

properties of parameters like the bias of a parameter the consistency and the efficiency and this is again tied back to the gas Mark of theorem uh also the

understanding of confidence interal will be really important in your career in the field of science and AI the idea of 95% confidence interval how it's

calculated what is this idea of um calculating this interval the lower bound and the upper bound what it means another very important topic from statistics is this idea of hypothesis

testing why we need hypothesis testing the idea of null um uh hypothesis the alternative hypothesis how you set up these experiments why it is important

why we even need it the concept of statistic iCal significance is very important how to calculate type one error type two error what is the difference between them what is false

positive what is false negative uh the statistical test like the student T Test the F test Anova test uh the uh two

sample T Test the two sample normal test there are so many test that um it would that um can be studied in this field of

Statistics but there are a couple of of them that I uh selected and um I would also provide you the links to that and you can also check them out and I would

highly suggest you to study them also this concept of P value is um very uh essential uh also this uh calculation of the P value how you can use it how to

interpret it its limitations and also this concept of inferential statistics so blows like the central limit theorem the of large

numbers how it is used when it comes to this uh experiments and this is tied back to the uh normal distribution function one of the most INF famous distribution function that you must know

as an AI engineer next up we have the dimension reduction techniques like the principal component analysis or the factor analysis and you can also add

here the panical correlation nysis so a CCA so if you are looking for that one place that in organized way can help you to refresh your memory or to study all

this in one place then you can also check out our fundamentals to statistics course because we are covering there all these different topics which is a prerequisite and it's a must for you to

know before you get into the next level in your AI engineering Journey so once you're comfortable with the mathematics and statistics you are ready to move on to the next step in your journey of

becoming an AI engineer the next skill set is the skills of data science so as an AI engineer you really need to have a good data science skills without good

data and without understanding whether you even have a good data or not and applying your data science skills um any of other skills won't matter because um

it's this phrase that is really uh easy to remember you can have a great AI model but if you put a garbage in you will get a garbage out and that uh what

you put into your AI model is your data if your data is a trashy is a bad data and sometimes you don't even know that you are dealing with B data because you don't have the data science skills then

it doesn't matter how much effort or how much money you will put in your um AI model how much gpus you will use or um how big your data will be if your data

quality is a bad one to understand these data skills you will need to have a data science skills so what I mean by that so

when it comes to um AI models they like to work and they are performing good if they are dealing with the clean data your AI models also need to use a

meaningful data a relevant one and also as an AI engineer you are responsible for the um for the ethical side of your

model and for that your data should be uh unbiased as well so um as an AI engineer you will need to understand how to clean data how to Source data how to

collect it if you don't have an AI engineer next to you and also how to pre-process data and here I mean identifying the uh Missing data in your

database to understand what is the mechanism behind it is it missing a trandom is is it missing not a trandom because this will then define whether you can impute the data so you can fill

in this missing data what kind of techniques you can use to fill in this missing data or maybe to drop it all together to understand whether you have uh anomalies in your data outliers how

you can use statistical and other techniques to find this outliers in your data and to remove it or maybe adjust it this concept of normalization you will

need to have a good understanding how you can filter your data how you can um group your data um tell story about your data before you even get into the model

development section and how to uh split your data to have the skills of um following the cycle of data preparation

data evaluation and also using the data as an input for your model whether it's a machine learning deep learning or an advanced generative AI model also

understanding how to uh visualize your data is really important as a data scientist you usually learn the um exploratory data analysis and how you

can use these different tools includ including Python and simple libraries like Seaburn and metli to visualize your data and as a data science skill uh this

is a must to also identify outliers to identify certain Trends and also to tell a story about your data so this is the

basically the pre-work that you need before you get into any moral development if you want to do everything properly and as a professional you also need to understand uh Fe engineering

skills which also is a data science skill so understanding how you can create new variables so sometimes for example you have multiple variables but

it's not good enough because you just need one and it's usually a combination of this multiple variables and by understanding how you can combine different variables in your database in

one place and uh create one single variable is what we are referring as a feature engineering so you engineer the features that then you can use as an input to your machine learning or your

deep learning or your AI model in general so this is about the data science knowing data signs uh will be um

will set you for Success when it comes to AI engineering career next up is the infamous traditional machine learning so

without understanding traditional machine learning there is no way to beable arounded AI engineer um if you don't want to be in this position where for every single problem

you use neural networks use you waste your company's money on the gpus or uh you spend a lot of time on using complex models that while you can use a simple

machine learning models if you don't understand this then you can never become this AI engineer that uh looks at problems not just from a research perspective but also from business or

Enterprise perspective so um that's why I always suggest to First Master the traditional machine learning and then only get into the next

point so here what I mean by traditional machine learning I mean to um understand this concept of classification regression supervised learning

unsupervised learning these different algorithms that fall under these categories like uh linear regression logistic regression decision trees uh

bagging boosting XG boost uh light GBM GBM and uh many other models including unsupervised models like K means hierarchy Cloud string or DB scan in

which cases which of your models you can use the idea is that once a PM or a business leader comes to you and tells you this vague business problem you as

an AI engineer you will need to quickly uh be able to figure out whether you are dealing with a classification problem regression problem maybe an unsupervised

learning program and you will also need to have this uh quick understanding okay I'm going to use most likely this models

in order to solve that problem and being able to understand this will be really important before you move on to any advanced moral uh

studying so um Beyond understanding the algorithms and if I believe if I remember correctly those are about 23 or 24 algorithms from traditional machine learning understand their mathematics

behind the statistics behind it what are their benefits what are their disadvantages because in each of these categories you also need to understand how each of these models work and um

have this understanding that for this type of problems for example when you have a lot of missing data you can use that model because it's more stable or if you are dealing with a data that follows normal distribution then you

will then you can better use another type of model cuz for each of this classification regression or other type of problems you will have many options

and it's up to you as an AI engineer to short list them and also from that to filter out which one you will use so beside this you also need to

understand how you can evaluate a tradition machine learning model what is this common cycle of the training testing validation what are these different sampling techniques or

resampling techniques uh what is bootstrapping what is cross viation what is kold cross viation or leave one out cross viation and also to understand what are the different evaluation

metrics depending on your problem you can use in order to evaluate your model for example what is the difference between using the mean absolute um error

versus the mean squared error in which cases you can use which one or are or the root mean squared error or um how you can evaluate a model that is in the

field of classification it is the F1 score um or it's the fbaa score which is more General version of the F1 score should you use recall should you pay more attention to the Precision

Etc so uh understanding also when to use machine learning when to use uh just rule based approach will be also important for you as an AI engineer so

um that is about machine learning if you want to uh Master the field of machine learning and everything that I just mentioned in one place you can also check out our fundamental to machine

learning course where we cover everything that you must know in order to become a well-rounded machine learning specialist you can also get a

certification from lunatech Once you complete your machine learning course so once you are comfortable with mathematics statistics and the traditional machine learning next up is

studying the Deep learning deep learning is at the heart of the Modern Art artificial intelligence especially when it comes to generative AI so all these

different Cutting Edge tools like the chat gbt The Dol Sora or the um different applications the um self

driving cars the uh robots humanik robots they are all based on narrow networks and narrow networks is this fundamental part when it comes to deep

learning think of the deep learning as more advanced machine learning where the models are able to study better uh with

a larger amount of data and this big data that uh the size of which increased more and more in the last decade made the evolution of the deep learning more

possible so when it comes to the Deep learning what I mean exactly is that you need to understand how the Deep learning differs from the traditional machine learning you need to understand the

architecture of neural networks uh and how it works the concept of neurons the perceptor this uh um in a simple way to be able to understand the structure of

neural networks the activation functions what it means this difference between different activation functions um and also understand in which cases to use

what this idea of hidden layers input layer output layer um how they are related to the performance of neural network um you also need to understand

the concept of for forward PA backward pass the idea of B propagation what the B propagation algorithm does the idea of loss function how you can calculate the

loss function for a neural network also how the training of neural network works so how it starts from the input then it goes to the forward path then does the uh the loss calculation the back

propagation Etc and also what is this idea behind it and how using each of these different making each of these different decisions like the activation

function or the uh different optimization algorithms how it will be impacting the performance of your deep learning model also understanding the different optimization algorithms like

the gradient descent stochastic gradient descent the RMS prop uh the momentum SGD Etc and of course the Adam or the adamw

these different algorithms will be really important for you to understand how the Deep learning models are being trained and optimize uh beside that you also need to understand the concept of Ving radiant

problem the exploring radiant problem um also understand um this different um computational graphs that are being used

in order to represent uh NE networks um also um how you can evaluate the performance of neural networks how you can use the cross entropy um and um

being able to understand these different um optimization technique makes the concept of mini badge gradient descent is also important and the difference between bch gradient descent mini BGE

gradient descent stochastic um gradient descent uh understand the concept of Haitian uh why Haitian is is being used what it means

to have a faster versus better performing neural network um understand also this batch normalization layer normalization what is the difference

beside between them understand the concept of residual connections and also what is uh gradient clipping cavier initialization basically how you can

initialize your neural network models of course um when I meant the fundamentals of neural networks I definitely meant also understanding what is the bias what is the weights uh what it means to train

a neuron Network the role of improving these weights and also you need to understand the ways you can solve these different problems like how to solve a Venum gradient problem how to solve an

exploding gradient problem um and also um these different techniques to combat the overfitting what it means to have an overfitting this comes from traditional machine learning but also in the Deep

learning it's still a problem and also understand how you can use drop out what is drop out uh what is this difference between random Forest versus drop out

what is the U L1 regularization L2 regularization versus Dropout the difference between them what is this idea of course of that dimensionality and what is this

difference between discriminative versus degenerative models what are out hand coders uh what is this idea of

reconstruction um error the um nonlinear counterpart of PCA basically when it comes to Outer encoders um understand

the architecture Behind these different sorts of deep learning models the um

Infamous uh Ann CNN RNN GNN grus lstms what is the difference between the lstm and RNN why they have been created what

are their problems what are their um good points and also how you can train these different algorithms and for what

kind of problems you can use what is p convolution what is pooling what is tried uh when it comes to a computer vision uh and in which cas cases you can

use what uh where you can use RNN where you can use CNN um and also the generative adversarial Network so we are getting more towards the generative AI

but I will tell you in a bit about that so this is basically about the theory when it comes to deep learning everything that I just mentioned and having these fundamentals in place will

be really important for you in order to go on to the next step which is understanding how you can use the traditional machine learning and traditional deep learning in the real

case problems which means uh learning Python and learning Advanced python um the AI Frameworks like the py Tor and tensor flow but not just that also to

understand a basic data structures and algorithms in Python and um beyond the python for data science also to understand how you can train a machine

learning model in Python how you can train a deep learning model in Python uh how you can uh do a visualization in python or how you can filter your data

prepare your data clean it basically everything where it comes to data science machine learning and deep learning its practical implementation is happening in a programming language and that's where the python one of the most

popular programming languages come in handy and my suggestion would be to learn next the python to understand how

you can um uh create uh lists variables how you can load data different sorts of data whether those are images text audio

how to work with dates how to filter your data how to group your data how to visualize it training a machine learning model training um deep learning model

how to make use of uh pytorch which is a deep learning framework in python as well as tanor flow this idea of tensors and how you can be Feld you can train

and deploy machine learning and deep learning model using python if you are looking for that one place to learn all what I just described make sure to check

the lunch. page because we have an

the lunch. page because we have an entire course when it comes to python for data science which would be a great starting point for you to Learn Python specifically for AI and data science and

also get a certification in that field and if you want to H check out all these different Topics in one place when it comes to learning make sure to check our deep learning in preparation course

which covers all these topics that I just mentioned in a unique Q&A way all right so once you have done that you have your mathematics statistics machine

learning deep learning and python as well as advanced python in one place you are ready to move on to the next step and here of course I'm talking about the

generative AI so when it comes to generative AI this by the way also includes the large language models here is what I would suggest you to learn in

order to call yourself a worldclass AI engineer generative AI is one of the most in demand skill set when it comes

to the 2024 and 2025 in this AI race in the companies like open AI the entropic Google meta Tesla xai all these

different um companies that are at the Forefront of AI Revolution they all are based on these different usages of generative AI CH GPT is based on

generative AI the Del is based on generative Ai and um the gini the perplexity the autonomous cars these

days all of the cutting get AI Tools in one way or the other are based on generative AI That's why if you are an AI engineer with the specialization in

generative AI this will be your year all right so let's now talk about what exactly it takes to become a generative AI based AI engineer so first up you

need to understand the AI foundations and you need to understand um where you can apply generative AI before you get into the theoretical part so

understanding also the moral development cycle when it comes to generative Ai and training techniques will be really important because different type of generative AI Foundation models can be

trained and optimized in different ways understanding different foundational generative AI models will also be very important think of like the generative

adval networks this concept of the um G and D uh so basically the generator discriminator what are their roles what they are doing how the Game Theory comes

here in place the um where you can apply generative aterial networks the mathematics and statistics behind it the mean Max Theory the Nasha equilibri room and also this concept of uh mode

collapse which is a problem for generative adversarial networks and how you can train and optimize these gens in order to create new data for example

synthetic data Etc so the next topic that I would highly suggest you to study is this variational out and coders this is yet another Foundation generative AI

model that I would suggest you to study to understand its mathematics behind it its uh statistics behind it the um idea of this difference between plain out

hand coders and variational outand coders which is a matter of difference between discriminative and generative uh type of models architecture behind it

the KO Divergence the elbow how you can train a variational out and code or this um all the way to the idea of reparameterization trick um and how you

can apply variation out encoder in practice of course next topic that I would suggest you to study is of course

the Transformers so Transformers they are the um mother of all the uh current Cutting Edge large language models when

you hear about Lama from met when you hear about gpts from open AI or the cloud Sonet from anthropic all these different large language models at the

base of them is the Transformers Transformers are at the heart of large language models and no knowing their history their evolution understanding what is the difference between

Transformers versus the rnns and stms from Deep learning understanding what is that the what was the reason of them being invented compared to the RNN and stms the idea of embeddings the

positional encodings how you can calculate them the attention mechanism the self attention mechanism the curies keys and values the single-headed tension multi-headed tension

understanding this entire AR architecture of Transformers step by step the mathematics and statistics behind it how you can calculate these different parts how you go from the input embeddings positional encodings

all the way to the logits which are the outputs of the Transformers will be really important for you as an innovator to understand because then you will be

able to understand this more uh variated versions of the Transformers because GPT models from open AI which power the chat GPT for example are A variation of

Transformer model and given that this uh modern models like llama the gpts or other ones um they are never published in its entirety so you won't know for

example what is the real architecture of GPT 403 uh which some call it also the first AGI model um for that I would highly

suggest you to go to the basics and to instead understand the Transformers once you understand the Transformers you also understand the cycle of pre-training so

what it means to pre-train a Transformer model do it by hand from scratch instead of just using um pytorch Library this will help you to completely understand

this um foundational language model once you are done with your Transformers I would highly suggest you to uh get into

the uh next topic which is the large language models when it comes to large language models uh this is yet uh

entirely a different um Universe um and for understanding this and calling yourself an expert in LMS uh you have quite a journey to go through but I will

make your life a bit easier and I will give you the step-by-step process and a specific skill set that you can um learn

in order to um Master the field of large language models so you can start with understanding uh what are the language models this idea of engrams the the

concept of predicting the next word and then um how this um large language models have been um evolved over time um

what is their unique sites um also understand what are the key large language models like I just mentioned the gpts from open AI the Llama from

meta the Falcon the bird from Google and also the Gemma from Google um the cloud Sonet from entropic these are different

uh very popular open-sourced or close- sourced large language models that you can use as part of your AI engineering journey and understanding these differences understanding how it relates

back to the Transformer models will be foundational for you also understand this concept of Open Source and closed Source what are these different tools that you can use when it comes to open

source of course I'm referring to the hugging phase and um other platforms that you can use in order to um use large language models and generative AI

in a more efficient and cost uh um efficient way so understanding the uh foundations of large language models will be really important for you like

the attention mechanisms that I just mentioned the language models the engrams uh the architectural Transformers the type of architecture you um you will meet in the field of LMS

like the encoder base or decoder base or a combination of them um the concept of tokenization from NLP the embeddings how you can calculate each of these parts of

the Transformers if you want to be someone who understands these large language models you want to know how to uh tweak them how to edit them and also

um in an intelligent way to use them in your job or create a platform so um you also need to uh be aware how you can prepare your data for

your large language model application like with the machine learning or deep learning if you don't have the data science skills then it's just a matter of wasting your time and money because

then uh this concept of garbage in and garbage out still holds um even if you have the most Cutting Edge LM model in front of you if you don't know how to

clean your data how to deal with with unstructured data how to prepare it and then injust it into this AI model then

this AI model will be uh performing poly that's why understanding this uh prompt uh templates this different type of models these different structures will

be uh Paramount for you next up I would suggest you understand the cycle of pre-training large language Model fine-tuning A large language model prompt engineering reinforcement

learning evaluating and optimizing in it so understand what it means to pre-train a large language model how do those large companies use um masked language modeling or Auto regressive language

modeling in order to pre-train large language model and then uh maybe do an example uh do a pre-training from scratch uh in order to get a taste of what it's like to pre-train a large

language model what are the different scaling clows what is the output of pre-trained large language model and what are the different Downs stream tasks that you need to be familiar when

it comes to large language models because large language models can be used for classification it can be um for uh predicting the next word uh for different tasks instruction based

approach for each of them the downstream tasks is different which means that the model and the way the model is trained is also different and also when it comes

to the large language models you need to definitely know how to fine tune a large language model so fine-tuning on a single task fine tuning on a multitask

instruction model um how to perform uh parameter efficient fine-tuning or the PFT and here of course I'm talking about understanding in detail the mathematics

the linear algebra behind Laura the chlora so the quantized version of Laura how to prepare your data for finetuning what are the steps behind

fine F tuning what is reinforcement learning so on supervised fine tuning supervised fine tuning um and uh

definitely um also experience um when it comes to fine tuning can you for example fine tune large language model on your own do you know how to use these different models

Etc once you are done with the pre-training and fine-tuning I would also suggest you to understand and learn this idea of reinforcement learning with human feedback not too detail uh you can

start with the basics uh just to understand this uh concept of the rhf or the reinforcement learning with human capital and why is that that we use this

in order to make our moral smarter then uh once you are done with that I would highly suggest you also to understand uh this prompt engineering so what it's

like to work with language model and to make it smarter once it's already pre-trained and fine-tuned in here

um I would suggest you to um to look into the best practices for prompt engineering how to do effective prompting prompt optimization and also

how you can apply prompt engineering along with the fine-tuning in order to create AI agents once you are done with that I would suggest you next to go into the

topics overx so to understand what are the concepts of the retrieval argumented uh generators and systems the vector databases integrated DRS with Gen how to

fine-tune with retrieve data this concept of Lama index um agentic RS Etc so once you are done with this I would

suggest as a final last two topics for you the first one is the evaluation and optimization of large language models

the mastering of the LMS to understand how you can use uh quantization knowledge distillation

pruning how you can use um Alm Ops to not just um train LM but also productionize it using topics like um

tools like longchain flask Etc and also understand how you can evaluate a large language model there are different benchmarks different data sets that you can use in order to properly evaluate

and compare a specific large language model to other world known Universal large language models so so finally I would say as an AI engineer you are the

first person responsible for the ethical and safe creation and usage of your AI models that's why we all have a joint responsibility to understand the AI

ethics the principle of ethical AI the bias in AI the privacy and data security in AI to understand this AI act from Europe to understand the gdpr act and

then understand the the regulations and the governance so um I hope this didn't overwhelm you if you have a good guidance and if you um have everything

in one place it will take on average 3 to 6 months for you depending on where you are and whether you already have the prerequisites in order to become a world- class AI engineer so AI

engineering is all about solving real problems not just the theoretical knowledge being able to understand all the theory the foundational knowledge along with the implementation of each of

these different topics ICS in the reality will be really important for you to become a job ready professional a world class tenic AI engineer who knows

the foundations and the actual implementation by Bridging the Gap between research and the industry application for that you will need a

well-rounded comprehensive training as on one hand and also on the other hand the Practical implementation with projects with ready resume to be able to

start applying or create your own up CU it's all about innovation in this field and it takes a lot of effort and motivation to combine this different um

skill set in one place but this also comes with a reward on one hand yes you need to put a lot of effort in order to become an AI engineer but on the other hand this is one of the most rewarding

and most in demand careers for the next decade so if you are serious about becoming a world-class AI engineer then you can also Al apply to our AI

engineering boot camp at Lun Tech to get everything in one place not just surface level knowledge but the theory the actual projects that you can also put on your resume those are cutting Gage

projects and as a results you will be able to call yourself a real AI engineer and Lear that dream dream job or to

become a Founder for a unicorn