Gen AI Comprehensive Roadmap

Key Takeaways

Yes. Hey, uh what do you mean by a self-paced course? Well, the course material is available in the form of video lectures and there are multiple practice problems that come along as well. So you can access all of it anywhere anytime and they are available to you for lifetime. No way. Oh yes way. Don't delay and enroll now. Geeks learning together. Hi everyone. Good morning and sorry you're late. I hope I'm clearly audible and clearly visible. Can someone give me a confirmation? Am I clearly audible and visible so that we can start with the session now? Okay. I hope I'm clear. Yes. Okay. Right. Right. Right. So, we can start. So, guys, I'd be reading your comments on my phone. So, there'd be a little lag, okay? There'd be a little lag between me replying to your chats and me actually answering them because again, as I mentioned, there's a little lag. Without further ado, let me present the full screen. Yes. And let me share the screen as well so that we can start. Share screen. Yes. Yes. Perfect. Okay. So, in yesterday's class, we were also discussing about data science, right? The road map for data science. And I hope you were there in the yesterday's class, right? What we discussed, let's do a brief summary. We discussed about the entire road map. We discussed about what exactly the important steps and the key things that you need to learn to actually excel in data science. Right? Initially we talked about the entire timeline. We talked about the entire timeline that how it actually started. So to give you just a brief introduction I'll again revise those concepts. Around 1850s the idea of AI came into picture. Right? Around 1850s AI came into the picture. The idea of AI came into the picture with the idea that machines can get smart, right? The machines can get smarter. So we can implement those tasks to machines. We can give all those tasks to machines so that human do not have to work that much. Perfect. First idea around 1950 around 1900s still the idea of AI was continuing right with no solid grounds. Around 1950s we got our first solid ground. We implemented the machine learning models, the ML models. They worked on statistical machine learning things. So we understood that we use the mathematical concepts. If we train the model on those mathematical concepts, then it can actually understand all these things in a much more better manner and they can predict the value for us. Right? They can predict the value for us. That was the case that we thought around 1990s. Let's say around 1990s machine learning came into production. Around 1970s also Harvard created its world's first self-driving car right good morning Rohan. Good morning Rahul. So they created its first self-driving car and in there machine learning was used. Okay around 1990s as I mentioned machine learning was came into production. So we started using machine learning into several several models and we started implementing them in different different models and we tried to implement them as much as possible. Around 2006 deep learning came into production. Now if I talk about the era switch that happened from 1990s to 2006 one big factor was as we've talked yesterday as well that uh during that time period we had internet cafes emerging. So whenever we had to play some game, we had to operate on a machine, we had to uh use computers and play games or whatever you want to watch movies, we just went to those internet cafes, paid for an hourly rate and we utilized those things. By the later term 2006, what happened is smartphones came starting into production. Right? So not everyone but people started having these smartphones. Right? Now during that frame that was one of the thing that we switched from a different era of an internet no internet era to an internet era and we initially started having smartphones as well but during that time phase one or more thing happened that was we transition from traditional CPUs to GPUs right now GPUs are much more flesers and right now I even have TPUs right so these are the things that have transitioned perfect around 2010 we had big data around 2016 we got the first model of charge GPD but we heard about them in the covid right and then 2025 we have heard about deepseek and we have heard about kim.ai Yeah, that's the entire transition, right? That was the first thing that we learned. Second thing that we learned was the entire structure, right? So, you have AI and then we had ML and then we had deep learning and then we had generative AI. Now, today we're going to talk about this generative AI portion. Perfect. Without further ado, let's start with the slides. I hope the jo is high. I know it's a Monday morning. It's a Monday morning but still I hope the josh is high right. So a little introduction about me also I given an introduction yesterday as well. My name is Ashan Singh and our data science faculty here at Geeks for Geeks. Okay. And I have I've had multiple I have handled multiple batches both uh in the offline training as well as the online training as well as the selected uh student training some individual who are working professionals also I've trained them as well then I've collaborated with multiple company projects like Asenture Salesforce right and um that's my entire attitude so I've been working in this field for two and a half years half years I was as a data analyst and then for the rest I was a data scientist. So I've seen the transition that has happened between in this field. One thing about this field is that you need to be having an appetite to learn because again and again new things are going to emerge out in this technology. Right? So you should have an appetite to learn otherwise you're not going to survive in this industry. Very plain and simple. New technologies that are coming to picture you should have an idea and should have an appetite to learn them. Perfect. and try to implement those tools on your own. Now we'll talk about the introduction to machine learning in generative AI. Perfect. Now again as we talked about yesterday, we talked about this entire thing. So this is the concept of artificial intelligence. Then further we had machine learning. Inside that we had deep learning as well. Right? And then we had data science. So data science is a subset of AI. As I mentioned these three portions are focusing on actually creating the things right. These three are actually focused on creating these new things right whereas the field of data science and all they are more focused for utilizing these things. So whatever they have already built they they're just utilizing these things. The data science is actually utilizing these things. It's a subset. Whatever model they have built in machine learning and deep learning, we directly just import them and use them as a data set and they will have help you to train and test the data and do all the functionality that is possible. Perfect. Inside it there is generative AI. Okay. Now, first of all, what is the difference between your traditional AI and your generative AI? Do you think do do you know the answers to it? What's the difference between your traditional AI and your generative AI? Can someone give me an answer in the chat box? What's the difference between traditional AI and your normal chat bots? Looking at your chats, guys, give me the difference between what's the difference between traditional chat bots, traditional AI and your generative AI. What's the difference between them? Anyone okay? Someone has replied AI is traditional and genai is new. Okay, but that's not the clear distinction. Generative AI can give solutions based on the detail you provided to it. Okay, that's also one of the answers. I'll give you the perfect answer for it and that's very simple to understand. AI can predict, genai can generate. That's the entire difference altogether. So AI can predict, genai can generate. That's the only source of differentiation and distinction between them. Right? So your traditional AI algorithms whatever they were doing, they were learning from the data set and based on that. Yeah. Yeah. It is in the back of my G. So it's there in the what's happening is that these traditional AI models, they already had these things. they already had these kind of models and uh we were training them. So based on that we are training them and they're giving a certain output. Now I'm giving a totally new entry. Based on that they are predicting that this could be the possible answer your traditional AI. Whereas if you're talking about the generative AI they are therefore generating the new content all together. Right? Moving on. So let's talk an introduction to Genai. Genai is in artificial intelligence capable of generating new text. Now it is generating new text. It is generating new videos. It is generating new images all together that's not even existing. Okay, that's the concept of generative AI. Moving further, what is happening here is nowadays what is happening? We are bound to LLM models. We are we are addicted to these LLM models. If I'm talking about chart GPD, you all are addicted to chart GBD. you have even the slightest of the problem you're just going to charge and entering your inputs and charge is there to help you out right that's the thing so we are bound to these LLM models now what is this LLM you'd say LLMs are large language models these chart GPT and all these things work on this generative AI model and these are your LLM models that are large language model that are going to reply to you based on the problem statement okay Now what's happening is if I again go back to my traditional methods this is your traditional methodology that we are following one is pneumonia and no pneumonia right this is your traditional one and zero zero and one pneumonia no pneumonia diabetes no diabetes kind of prediction algorithms right this is your initial AI models your AI models normal AI models that are used to predict the Now what is happening here is you either have pneumonia or you don't have pneumonia. How it happens is I give 10,000 images of no uh I give 10,000 images of pneumonia patients. based on that it learns that okay because of this and this or whatever it is using whatever metrics that we are using here to determine that it has pneumonia based on that suppose I'm saying because of this this this and this riptage the person can be determined that it has pneumonia right and based on these these these suppose these are your features these these these features you can understand that the person is suffering from pneumonia then what is happening is I'm giving 10,000 images of no pneumonia patient that are completely healthy. Again, it has c certain photos. So now it can actually understand that okay because of these things that that I was taking that was wrong because of only these two parameters a person has pneumonia or it does not have something like that goes behind the algorithm. Now that's not the perfect example to it. That's just an intuitive way of telling you to how exactly you can understand that what exactly is the mind that goes behind these kind of algorithms. So based on that it can easily understand that yeah okay this guy has pneumonia and this guy does not have pneumonia. Perfect. Now these are your basic projects that you build when you first learn about these uh models. Right? So pneumonia no pneumonia diabetes no diabetes and uh there were few more elements to it. So just one and zero kind of problems. Yes and no problems. Okay these are your basic problems. Moving on again what is happening here is that you have Gemini as well. You have Gemini, Chad, GVT, Knock and you have different different kind of models. You have uh BERT, you have um what are different different you have deepse.AI all these models they there are for you what they are doing is they are just doing anything and everything that we are asking you for. So here I've given a prompt that suppose I want to mail something I want to mail to an individual that uh this based on this certain pattern I want to create a mail from them and I have to send this mail to suppose my manager or my uh my colleague or to to an HR or wherever it is whatever mail I'm asking it to generate it is generating to me on the go. What is it doing? It is generating new content on the scratch from the scratch based certain references that it has taken in the past. This is your generative air. It is generating new content all together. Right. Next, this is the image generation. So what is happening here is Hulk meditating. So I've given it a prompt that Hulk is meditating. Now this is the kind of new images that are being generated right we don't have this idea altogether if I give this task to a graphic designer he'll have to research it for a few days that okay this is how Hulk looks like this is how people look like in meditation so when I've given it a prompt like Hulk meditating it is understanding that it has to be in a serene place so it is showing waterfall and all it's in a green place and Hulk is sitting there's a certain posture that you sit in when you are meditating right You hold your legs all together and then you have your arms on your for on you have your arms on you have your hands on your arm on your legs and all these things. So it is understanding all these things and it is creating that on the go. right now if I again just a fun activity. Okay, just a fun activity. If I do this, this is your chart GPD. Perfect. And if I go to the model search of if I search for different GPD models here, GPD 4.5 as well that is emerging out. So this is the perfect version right now. So this is the uh premium version of it. Let me just explore GPTs and that I want to explore which one I want. I want Dalai. Dalai is initial model of charge to actually build these kind of things. Now give me the wildest prompt that you have. Give me the wildest prompt that you can think of. Any prompt like just give me key elements of it and I'll start making the prompt. Anything out of the blue just keep it on. Just keep it censored as well. Okay. Give me the wildest prompt in your chat box. Waiting for your chats. image of a zebra. Okay. Create me an image of a zebra. Anything else? This is very pain. Okay. Running chai. Okay. Sitting on top of elephant. Okay. Sitting on top of an elephant. Running chai. Okay. Let's take sipping chai or Yeah. Sipping G. Give me something more. Something more in space. Okay. In space. Anything else? Elephant eat sugarce. This is just an abstract idea. You can give any prompt that you want and just we are creating this visual out of the blue. Okay. in space. On Saturn. Okay. In space on Saturn. Let me add some new new elements from my site as well. uh on Saturn looking at Earth right from traveling in space who text okay where right next do it I don't think Donald if I give the example as Trump it will it will just it will give me a censorship because of that okay so I'll just give it like A person is [Music] traveling in a space x in a space in a space space suit. I'll say again spacex is space suit in space. Right? This is a very precise prompt. If I give an output to it, it has certain elements like create an image of a zebra in an alternative universe. Okay. Is it is it perfect according to the prompt that I gave? Create an image of a zebra sitting on top of an elephant. Zebra sitting on the top of an elephant. Sipping tea in space. Okay. Sipping tea in space. That is also being given. Right. And then looking at Earth. Okay. I am assuming this is Earth. This is Saturn. Right next to it. Uh where right next to it is a person is traveling a space suit. So there's a person traveling with them. Right next to them in space. This is a perfect image. You can say that yeah these are your perfect images that it has created. Now if I ask it to I can give another prompt as well like uh create it in anime style. I can change the style as well. So what you are doing right now is you were creating these Gibli style images, right? So you can change the type of it as well. So created an anime style. So animation style I'm creating that now right. So this is the animated style it has given you. You can create the Giblly style as well. You can add abstract. You can add uh 35 cubism dawn and all these things. Now the idea that goes with it is that if I create a similar kind looking at the yeah perfect right. So the idea that goes behind is that if I asked a normal person or I as a graphic designer to create this kind of image, they will have to take references from multiple places. They'll have to adjust it accordingly that it is in space, it is in Saturn and all these things, right? It has to they have to install it in that that place. one thing they have to research a lot and all these things have to be done and they'll take certain amount of time for it as well. they'll take suppose uh 5 days to 6 days to actually work on it right and the team would be working on it then only the certain individual can do the certain thing but what is happening here is within seconds I gave the prompt and then within seconds it has created this entire thing for me right so that is the thing okay that is the entire thing that we are following now if you if you have the idea of it what was happening was you guys were creating these Giblly And from the director of the Gibli images he was giving a statement like this is the extinction of art and the art is being not respected and all these things right we have all seen that so this is the case that is happening but let's move past that so in summary what is generative AI generative AI is a field that is able to generate new content altogether perfect it is able to generate new content all together for us Right. Okay. Now let's check the evolution of GI. Okay. When talking about the evolution of GI, you'll already have certain idea that what exactly is a thing because in the earlier modules we have already talked about it. Okay, we've already talked about how it actually work. Now the first thing is that we had the statistical machine learning. Okay, this was the fit simple features where you have an area and you have a price. You have an area and you have a price. Based on that, it can actually predict that yeah for 3,400 area this is the price. For this area, this is the price. So, it understood the relation between area and price. Now, based on that, if I ask to give an area of 9,000, it can predict me that this would be the value that there would be existing. Done. After that, we transition to neural networks. Now in this neural network what is happening is here we actually had the GPUs and all this is a portion of deep learning right so it has if I'm given 10,000 images of a cat so it is understanding that yeah this is a cat based on its whiskers based on its ears based on its nose legs and all these features it is passing that through several several layers and finally it is learning that yeah this is a cat similarly I'm giving 10,000 images of a dog as well okay again through through its tongue, through its eyes, through its ears, through its whiskers, legs, tail, it is understanding that yeah this is a dog 10,000 images and based on that it is learning all these things. So once it has learned all these things then it can even bifurcate that yeah this is a cat and this is a dog itself all these things are done. So these are your complex feature your complex problem statement that is being done in very easy in this manner. Perfect. Next if I talk about your models right this is the thing that we have done initially we talked about statistical machine learning that was your basic simple step steps then you have your neural networks right that was your deep learning portion neural networks further after that we transition to recurrent neural networks now what exactly is recurrent neural networks let's try to understand that as well what is happening here is I'm translating it. So from Spanish to English it is writing you are very well and suddenly it is writing you are very baby shallow right again you are very well and suddenly it is writing baby shadow and we have seen these kind of things happen in real life with us as well right so what's the general idea that goes behind it what's the general idea that goes behind it this is where the concept of RNN actually takes place okay this is the concept where RNN takes place now recurrent neural What exactly is this? Let's try to understand that as well. This is the concept of RNN. The concept of RNN is that RNN they actually they transition into the learning phase as we are learning right what they do is right next right before what they were learning they they're not not going to forget all these things to forget the memory of the past. No, they just remember that memory and whatever they have learned right now and whatever they already have memory of the what they learned in the last time they try to merge it together. They try to merge it together. Okay, they try to merge it together. So once it is being done in that manner, it can merge these things together. It tries to indulge both the technologies together and tries to make a new reasonable output for it. Right? So let's understand that in this case is Java preferable at the start of the journey in AI uh not exactly J I'll tell you about what exactly which programming language has to be done okay so an agent coming to picture these are the pillars and these pillars have certain information like this pillar has information of I this pillar has information love this pillar has information eating this pillar has information pasta now it is transition ition. So it goes to the first pillar and here it learns the word I. Then it transitions and here it learns the word love. Right? It is following in this pattern. So what is happening is initially it goes to this pillar. It learns the word I. Now it tries to make sense that what exactly is I? It's not understanding it. Right? No worries. It moves further in the picture. Here it learns the word love. Now it remembers the word I. It remembers the word love. It tries to make sense. I love I love something. what it's not understanding it so it's again storing that in the reserved place so I love is being stored further it moves in the picture and here it learns the word eating now it tries to make sense I love eating I love eating what what exactly is I love eating perfect right so it has that idea in its mind finally it moves towards the last pillar and learns about pasta from now it indulges all the idea and tries to mix them together so I love eating pasta. Now it is perfect. Now it understood the entire meaning of what they wanted to say that is I love eating pasta. Right? This is the concept of recurren networks. They don't forget the memory that they have. They just try to implement that memory and try to indulge it together. Perfect. If I show you again a simple translation. This is a translation from English to Hindi. Okay. And the and the title is I love you. Perfect. So in English it's I love you. And if I see that the entire translation in Hindi what is it? It's correct. So the translation of I love you in English is right. But if I see the literal translation that would be I would be me love would be you would be. So it would be but it is writing this is the concept of RNN this is the exact concept of RNN they understand the concept they understand the meaning and with each word they try to make sense as well so this is your language models as well these are your language models where actually interpretation of these things is very necessary got it now let's talk about autocomp completion how exactly autocomp completion And where exactly have you seen the concept of autocomp completion? You've seen the concept of autocomp completion in places like uh your Google when you're writing something or mail if you're writing something you have certain option to autocomplete it or on your LinkedIn if someone is messaging you out there right so you can directly message them back using this method that uh yeah this is the way of doing that right so in LinkedIn as well you have certain options that if someone writes hi you have options like hi Hey, where are you? What are you doing? I'm fine. Something like this. Right. So, you have these options right next to it. Got it. So, let's see how exactly auto completion also work. So, suppose you get a message from someone that hey, we have a potential collaboration opportunity. Do you have time to talk? You're double, right? And you're replying to Angela. Angela, thanks for and then autocomp completion comes in picture. Now auto completion we'll have certain options like Angela thanks for reaching out to me contacting me giving me the opportunity and you're a pickle. These are the four options you have right now. Okay which option are you going to pick? Obviously you're not going to pick you are a pickle right because no one says that. It's just that in the top three it's something that you're going to use either reaching out to me, contacting me, giving me the information, giving me the opportunity or you are picking. Now there are not only four options, there are n number of options present there. Okay, it's just that right now four are being presented to you. Got it? So with each number with each answer there's a certain numerical value placed in front of it. This is your weights. This is the weights that are associated in front of him. So the higher the value of the weight, the closer it is to one, the most probable it is the answer. So Angela, thanks for reaching out to me has the is the most highest weight. So it would be presenter right next to thanks for contacting me is also very good. So it is 0.90 giving me the opportunity 0.87 and your pickle is 0.20 because no one replies that right. It start people start using me contacting me more Angela thanks for contacting me the weight associated with it the numerical value would be increased and if it's increased greater than 0.97 then it would be the auto completion that we get that's the concept of autocomp completion okay so language model is an AI model which can predict the next word or a set of words with a given sequence of words I told you this before as well now let's talk about this GPT4 4 GPD4 has 175 billion. Okay, this is the input layer. This is the hidden layer and this is the output layer. Right? The input layer has certain things like weight and height. Okay? So what is happening is based on this weight and height parameter I'm predicting the gender of an individual. Perfect. Now I'm taking this weights. Okay. Weight one weight and height. And based on that I'm passing through certain weights. through these weights W1, W3, W2, 3, 4, 5, 6. I'm passing these values in the form of weights to these hidden layers. Now obviously they would have a data set that already has the information for input layer for taking the input of weight and height and based on that it is predicting the gender. There's already a data set that is working on it. The models are already there and they are actually giving you the actionable output. It's just that all of these is being done in the hidden layers. That's the reason the weight the input layer the input is being transferred through different weights to the hidden layers and in the hidden layers they are doing all the computation and further through these parameters W5 and W6 they are getting to the output layer and predicting as the gender. So through these weights W1 W2 3 4 these are different little different parameters weights that are transferring this to the hidden layer. Ed layer is doing all the computation through W5 and W6 parameter we're getting the output that is the gender of the individual. Now for this particular good piece of case we had six parameters six weights that are being used. W1 2 3 4 5 and six. Six parameters were used. Okay. And this is a pretty good solution. If I'm talking about GBD4, GBD4 has 175 billion parameters like this. 175 billion parameters. Just try to imagine the scale it is working on. how many uh input layers, how many hidden layers, how many output layers and what kind of relations that they're developing. Right? This is GBD4 your transition from GPD4 to GP 40. 40 has 100 around 230 billion parameters and we have 4.5 I don't know the parameter limits for that as well. So that's the reason these things are getting more and more complex. The more and more complex that they are getting, the more faster and more accurate that they are getting. Perfect. Now let's talk about the final stage that we have reached. You have your basic statistical machine learning. Then you actually transition into neural networks. These are your neural networks. Okay, that we have learned. Then we transition to recurrent neural networks. You have different different recurrent neural networks that we talked. And finally you have transformers. Now transformer is the most advanced stage that we have learned up to now. Got it? So, statistical machine learning, neural network, recurrent neural network, and finally you have transformers. Now, it's quiz time. What is the full form of GPD? Give me the answers. What is the full form of GPD? You all have used chat GPD. You all have used that multiple times. Just give me the answer. What is the full form of GPD? Okay, once again there. One second. Give me the answers in the chat box. Just a second. So what exactly is the So what exactly is the full form of GPT? Write that in the chat box. What is the full form of GPD? Yes. What's the full for GPT generate performing transformer? No. Generative pre-train transformation. Okay, let's check the answers. Generative pre-train transformation. Yes, that's a perfect example of the answers that is generative pre-trained transformers. That is the example of GPT. So what it is? It is generative pre-trained transformer. So it is a transformer that is pre-trained and is there to generate these output for you. Right? This is the example of a GPT full form of GPT. Next, next let's talk about the different kind of transformers you have. You have text to text transformer, you have text to image transformer and you have text to video transformer. Right? So if you have the text to text transformers, what exactly is a text to text transformer you're giving it a textual input and you're giving it you're getting a textual output from it. Examples like bird and GPT are there to give you the exact same thing. The texture text transformers is bird and GPT. asking it a prompt like create me an article on Republic Day or create me an article on perfumes or create me an article on spices right and you're giving a textual input and you're getting a textual output for the same as well that entire text format that this is article and all these things so you have different different transformer like this you have bird you have GPT right then you have text to image transformers we just use text to image transformer right now one was DLE One was stable diffusion. These are your image creation models. So if you give a prompt like we gave a prompt that create me an image of an elephant sitting on top of a zebra sipping tea in space and all these things, right? You're giving a textual input and based on that you're getting an image output for the scene. That's a text to image transform. You have DLE, you have stable diffusion, you have midjourney. Even in your phones you have uh meta right everyone has their phone you have WhatsApp in your WhatsApp you have meta right so meta is also working on llama 3.2 model 3.3 model it is also an image creation model right so you can create images from that as well just try it out that's in homework for you okay you have llama 3.2 do you have midjourney, you have bird, you have dallay, you have stable diffusion, you have knock, you have different different models. So you can create these images and finally you have text to video transform. That means if you're given a textual input, you're given textual prompt and based on that a video output is being created like create me an image of um of a person running in the beach running towards the ocean. There's a girl that is running in front of her. The sky is blue and uh it's the birds are flying ahead. It's a beach house and they look happy. It's kind of like a romantic ending to the scene. Right? So I've given it entire scenario output for the scene. That is the example of a text to video transform. The example is open AI Sora. There's OpenAI Sora and that can create this entire thing for you. Got it? Next. Now after learning all of this, you must be like, okay, is there any money in this field? Is there any money in this field or not? Okay. I'll tell you the simple answer that yes, there is money. There is money in this field. Okay. Okay. Then you must be like that what exactly is the road map? Can you give me the road map? Before I tell you the road map, I'll tell you the condition as well that is happening. Anywhere there there is data. There are going to be individuals for data analysts, data scientists and these genai developers. There are going to be need for those people. Now it's just that in India right now the data is being there. Right now in the past three four to five months I've seen a transition in India for the reference of data and the jobs for these what is happening is through these quick commerce industries like your uh Swiggy uh blanket Zeppto what they are doing is they're doing the quick commerce within 10 minutes you're getting delivered so you have they have a huge database and if you check the jobs for data analyst data scientists and gener developers you're going to see that these big companies are asking for data scientists And these roles opportunities are being opened in Zomato, Swiggy, Blinket kind of companies, right? Because they have huge amount of data and even you have your startups, right? These startups, they are also getting the data from somewhere and they're actually hiring these data analyst, data scientists to find out the output for them and they can utilize that output, right? So the thing is that yes we are needing these things and right now India it is at a slow pace but it is developing it is developing for the condition for the countries like Silicon Valley, USA and all these things already have these huge amount of data set and they already have these things deployed but in India it is transitioning. So I'd say this is a right time to actually get inside the industry around 5 to 6 years I don't know 5 to six is a very huge number around 1 to two years you're going to see the trends being increasing if you already have the solid ground right now then you can easily get transition to that field and you can get a huge amount of money there then this field is lucrative I'll tell you the road map the road map is first of all you'll have to learn a programming language and we we are very uh focused on Python programming language, right? You have Python, you have R. But again, Python is a programming language that we're going to use. Why so? Python is a Python has a rich library base. It is very easy to write and all these things. But majorly, Python has a huge library base. Through that library base, you can do the manipulation, you can do the adjustment, you can do the data cleaning, data visualization. All these things can be done easily through this. Got it? So basics of programming language in Python should be done. Then you need to know about the data structures in Python like uh list dictionary taple all these things you should know. Then you have certain libraries for manipulation of the data. You have pandas, you have numpy, you should know these libraries for data manipulation. And then you have visualization libraries like uh cbond, mattplot, you should learn about that. Flask is just for deployment. Okay. After this you have the basic machine learning algorithms. You should learn about basic stats. You should know about supervised, unsupervised, regression based. So basic statistics, statistics, regression problem, classification, what is naive base, what is KN&N, what is RF trees, clustering, DB scan, hierarchal scan. After that you should have the concepts of NLP also known. What exactly is NLP? Why NLP? one heart encoding bag of words TF word tove average word tove then you're going to have your basic deep learning knowledge what exactly is a neural network what is ENN what is forward propagation backward propagation activation function loss function optimizers what are all these things then you're going to combine NLP and deep learning together that is advanced NLP tutorial so RNN LSTM RNN GRD RNN bonction STM RNN encoder decoder attention to call sequence to sequence and finally you have transformers. Got it? If you're doing just these amount of things, you're good to go for a data scientist. You are good to go for a data scientist. Okay? And also these you have two specializations. You have NLP and OpenCV image computer vision. So you just have to specialize in one of them. If you want you can get a basic idea for both but specializing in one of them is good enough. Done. After this this through this you can develop as a data scientist. After this is your actual journey for generative AI. Okay. So you have to learn these things like open AI how to use open. What is lang chain? Okay. So all the LLM models are being created on these lang chain. You have lang chain lang lang graph. All these things are the ways through which your uh these models are being created. After that you have chain lit. It's where you can actually create these chat bots for your LLM models. Okay. Then you have Gemini. You should learn all these technologies. There are a lot of few technologies that I've not mentioned but these are the few things that you need to learn. Then you have vector database and vectors too. So different vector like fi chromb fs vector db lance db cassandra db and they're also dedicated for which one to use at what point. So if you have a dedicated vector database and you have an open source available. So these are your options. In this you have you might have a few names like post cogs you have heard the name of these right. So all these are storing for vector database and vector source. Now research over what exactly is a vector database and what exactly is a vector store. Okay. So data is being stored in the form of vectors and we actually have to develop on that. So what exactly is that? Understand that. And finally you have the deployment section. Deployment of your LLM projects. You can deploy that on AWS. You have Azure, Langmith, Langer. Then you have Hugging Pace. You have lot of options to actually do that. Now after this I'm open to questions. If you have any questions you can ask me. And even if we are not able to ask me all the questions today, I'll do one thing. I'll share you share with you my LinkedIn ID my LinkedIn ID so that you can connect with me there and we can further continue the discussion if you need any help with me. Okay. So this is my LinkedIn ID and in case you need this you can further ask your questions there as well in case your questions are not answered today. Okay. So if anyone has any questions you can ask me. Is it linked with DevOps? No, not exactly DevOps. It's a different thing. Not exactly with DevOps. No. In yesterday's session also I told you that you should have an idea of big data as well. Okay. So it's a good thing even that even if you're a data scientist you should have few things like you should learn about SQL. That means you should learn about how exactly you can manipulate a database. Database manipulation also should be known. Apart from that you should know what exactly is a big data as well. So technologies you have huge amount of data man. So technologies like Apache, Spark, Hadoop all these things should have you should have a general idea of them because that leaves a little edge for you. Do you need to learn C C++ Python web tech for coming in? No. For generative AI, you just need to learn about Python. Just need to learn about Python, that's you'll be good to go. Okay. If in case these technologies are integrated in the future, which is very unlikely right now because nowadays what is happening is that in case any new technology is coming, any new model is coming, the first thing that they are checking is is it compatible with Python or not. So, Python is already dominating in this market. Don't need C++, webdev and right now all these things. Could be that in the future if they integrate these things could be it's the case but right now no. Any other questions? Anyone? Any other questions? Road map for data analytics. Okay, if you want the road map for data analytics, that is very easy. So, uh, initially you start with Excel. Okay, initially you start with Excel, you start with Excel, then you follow a SQL. So, you learn about databases, then you start with Python. So in Python as well you have to figure out a lot of things like uh what is Python the basics of Python basic data structures of Python and then you further enhance your knowledge through library where you see what exactly is a library what are the different libraries like your M plot clip semon and all these different different libraries and finally you have a visualization tool like a PowerBI Tableau PI tools business intelligence tool so these are the few things that you need to know and after that if you're done with that you're good to go for data analyst DSS is needed for geni right now the thing is that if you go to the these big companies they are actually ask they have a round for DSA not these small companies but these big fan companies I'm talking about they are holding a u an interview round for DSA why exactly DSA DSA not for asking you that do you know about stack do you know about you no they're just checking your logical ability skills so because again in this field of data science and data analyst and genai. You should have an idea of what exactly uh you should have a good problem solving skills, good logical ability skills, right? They are checking that through the through your DSA rounds. Through your DSA rounds, they're giving you question and they're checking that if you have do you have a solid rigid one way of solving it or do you have multiple ways and are you clear on those concepts. So based on that they're checking your logical ability. Okay. uh Rashtri Bharaj okay so through that they are doing this so DSA is somewhat important if you're actually targeting these big fan companies mini sharma connection connecting for LinkedIn I am doing MCA in AI from LPU guide me for it I'm transitioning from sales marketing okay so you you are from LPU okay I'm actually traveling to LPU by for uh a hackathon there in 23rd or 24th. I hope I can meet you there. Okay. So um the transition would that you are if you are from the non techch background if you're not from the nontech background we need then you'll have to learn a few things you'll have to learn exactly the coding principles. If you have the idea of coding then that's good. So start with Python programming language. Okay. What exactly is Python programming language? As I mentioned study the entire structure of data analytics first that will help you develop your logical abilities for which kind of data set what kind of uh ideas should I go with that will help you build your logical ability. scope the road map of data analytics Excel SQL Python and Tableau PowerBI whatever is visualization tool first you can skip Excel if you want SQL Python and a visualization tool that's good enough for beginners even if you're transitioning from a non- tech background okay learn about this then you can further transition your knowledge for data science right what exactly is supervised learning what is unsupervised what are different different techniques that you have right all these things And further you can once you're done with the data science then you can actually transition into your portion of generative AI. Okay the entire road map is present. It's just that you should have an an appetite to learn and be continuous in this field because again in this video you're going to forget a lot of things if you don't remember it correctly. I hope that answered your question. Okay. So we're running a little uh late. Okay. Now, uh that's it for today. Okay. Yes. SQL Python visualization tools. First that we need and then actually go for learning the data science portion. What exactly supervised unsupervised and all these things this will help you build a strong fundamentals. Okay. And based on that you'll actually have no problem. You'll have little to no difficulty in your data science and further journeys. Okay. So that's it for today's session. Thank you everyone for joining and listening to my rant for the past 1 hour. Okay. And uh I hope you even if one of you individual have found something important that is very helpful for me. So I hope you had something you learned something from this. Okay. And I'll see you in the next live. Okay. And we'll pick up some again some interesting topic uh so that we can actually use that for our further things. So I'll see you then. Okay. with a new idea with a new topic and if I if I have continuous sessions we can start with a topic as well right so we need could be that I'm starting with a I'm not I'm telling not right now not sure about that I'll check my bandwidth if it's possible then I'll start with the Python basics I'll take an entire sessions of Python for free okay so it's plan for the best and I'll see you in the next live okay thank you everyone thank [Music] Heat. Heat. [Music]