Intelligent Tutoring Systems (ITS) leverage AI agents to provide personalized, scalable education by replacing human tutors with software programs that can adapt to individual learner needs. These systems operate through three main paradigms: Socratic tutoring (using thought-provoking questions to guide students to discover answers), gamification (integrating game mechanics like rewards and quests to increase engagement), and learn-by-teaching (where students explain concepts to the AI agent, reinforcing their own understanding). AI agents are software programs that autonomously perceive their environment, reason, plan, and take actions to achieve specific goals, utilizing Large Language Models (LLMs) as their 'brain' to break down complex tasks and utilize external tools. The key advantage of ITS is the potential to make quality education equitable and accessible to larger populations, particularly in underserved regions, by providing personalized tutoring at scale.
AI MASTER CLASS本站添加:
while the session the theory part is being conducted because we want you to preserve uh the battery in your laptops for the hands-on session if you do not have enough charge in your laptop at a later stage you won't be able to complete it and also I suggest students to work as pairs as they progress because you can then you know even if one laptop goes off you can still continue on the other one all right okay then I will uh end my talk here and then I want you to take forward I I invite professor Jagat onto the stage to start off with the session.
Okay. Good morning everyone.
Uh I'm Professor Chagat. So in the mean uh so the slide is loading. So in the meantime I'll just introduce myself. I hope I'm audible at the back.
Clear? Okay. Thank you. So I'm professor Jagat. I am a associate professor at Bitspilani Pilani campus and I'm also an academic mentor at Bitspilani digital.
And today I'm here with on behalf of Bitspilani digital. Bitsilani digital is an online business online programs division of Bitsani wherein we offer so many online degree programs and certificates.
So u I'll be talking about uh the intelligent ting systems today and how AI can be you know the power of AI can be used for you know intelligent tutoring a very powerful ting personalized learning. Okay. So let me get started once the school gets loaded.
By the way, I'm also from Vishakawa. I studied in Visag until my class 12 and after that I moved to Bits Pilani for my undergrad and since then I've been there I studied undergrad masters and PhD from BITS. Also I did a post talk at Northwest University in Chicago and then I came back as a full-time faculty at this time. So um I was asking sir if I can switch to Telu in between but sir was mentioning that there are some students uh who who may not understand Telu. I'll I'll keep it keep the session to English but you want to talk to me in Telugu I'll be more than happy because no I hardly get to talk to Tel in to students because in bitani it's like mixture most of them are not telu so most of the time I speak in English I'll be glad if somebody wants to interact with me one to one speaking mostly okay so we have this team here so let's get started okay so u yeah about my work that I do at BETS. I do research, I do teaching.
So I do a lot of research in uh in the area of uh human computer interaction.
Uh I also build uh AI tools, generative AI based tools specifically for education. And uh at Bits Digital, I also develop scalable and online uh scalable and equitable online programs.
We have like BSE and MSE programs that are completely online. Anyone can register to them. And additionally I teach at the Pilani campus and I also teach at Corsera. So my association with Corsera. So I have taught three courses on Corsera uh in the in our degree program of BSE computer science which we offer on Corsera and uh most importantly I also received a uh an global innovation award in 2023 uh you know for for the for one of the courses which I recorded for Gosser. as you can see. Sorry.
Okay. All right. Okay. I'll Okay. I hope I am still audible at the back.
I hope I'm still Okay. Great. Yeah. So, you can see uh in one of the pictures towards the right. Can you recognize the person standing next to me? Andrew Ng.
Yes. He's the he's the founder of Corsera, godfather of a IML education.
And I take lot of inspiration from him.
That's why I put the picture here. and uh I had the privilege to meet him at one of the Corsera events that happened in Las Vegas in 2024.
Okay. So let's get to our topic which is intelligent tutoring system. So uh Kushbu these are the top. All right.
Yes.
Okay. So let's get talk started with our talk and uh let's begin with what tutoring or learning paradigms are.
Okay. So, how many of you I mean that's a vague question I'm asking. So, all of you are students, all of you are attending classes. How many of you also have the opportunity to get tutored? You know there is difference between tutoring and attending classes, right?
How many of you have the opportunity to get tutored? Anyone would like to say something or do you have an opportunity to, you know, tutor someone? There are two questions. One is are you being tutored?
Second is are you tutoring someone?
Anyone?
No one.
One night before the exam, what happens in the host?
One good student will tutor 10 other students, right? That's what happens, right? Okay. So do you agree that that one night thing works better sometimes?
Why?
Because it is personalized. You know there is one person teaching sometimes 1:1 or 1 to 3 or 1 to 4.
That's all. So you know it's personalized and you know the way you learn is sometimes better. I'm not saying classroom teaching should not be there. Classroom teaching should be there and tutoring follows classroom teaching.
Okay. So there is difference between classroom teaching and tutoring. So tutoring has to follow classroom teaching. Okay.
Heat.
Heat.
Heat. Heat.
Heat.
Heat.
Heat. Heat.
Heat.
Heat.
If you're adding two integers, what do you think the return type should be?
>> An int, right? Two ins added together is still an int. So int.
>> Spot on. Put the name, the inputs, and the return together, and you get this.
Read that first line back to me in plain English. Don't rush it.
Okay. A function. Its name is add. It takes two ins and b. And the int at the front means it gives back an int. Did I get that right?
>> Every word of it.
And that return line, what's its job?
>> It sends the answer back to wherever the function was called from. So, whoever called add actually receives the sum.
>> Let's run it in your head. Walk me through what this program prints out loud, step by step.
>> Okay. Main calls add with five and three.
calls add with five and three. So inside add a becomes 5, b becomes 3, a + b is 8, it returns 8 and print f prints that so it prints 8. It prints eight.
>> That's it. And notice you just traced it yourself. No guessing. Those values five and three. What do we call them when we pass them in?
>> Arguments. The arguments going into the parameters A and B. I think I'm finally seeing how the two sides connect.
>> That's the whole idea. One definition written once, reused with different arguments as many times as you like.
Last one. In a single sentence, why do functions matter?
>> They let you write a piece of logic once, give it a name, and reuse it everywhere instead of repeating yourself and fixing the same thing in 10 places.
That actually makes sense now.
>> That's the heart of it. And you reasoned your way there yourself.
Next time, we'll look at passing data different ways and returning different types.
Nice work.
Okay, so that was a sample video obviously generated by AI not by me but my student. He was supposed to be here but he couldn't but yes. So that was giving a small demonstration of how Socritic tutoring looks like. So do you feel having such kind of tutoring helps you understand things better? Right. You can see that teacher is not giving answers directly rather teacher is asking thoughtprovoking questions and and teacher is trying to you know kind of ensure that student himself or herself actually comes up with the answers in that way the learning will be very very you know very effective. Okay.
So this is a tutoring paradigm. So we are not talking about classroom teaching. This is tutoring paradigm.
Okay. So that was so learning. Now similar to socritic learning in a in a different way there is something called active learning you know or you can say learning with gamification and rewards.
Do you play video games on phone or on TV?
Who played games on TV? That cassette old cassette contra Mario in recent days.
In recent days, yes, you still playing Mario and Contra.
These days it's all you know you play games on mobile. So do you play any games you know where you most games today you know when there is a task given you achieve the task you get some rewards right that's how games are so when that kind of a game wherein there are rewards for every task accomplished can keep you engaging can't we you know induct a similar technology or a similar uh way of uh you know doing things into learning teaching learning practices what do you Can we integrate games with learning?
Would that make it more interesting or not? Yes.
You didn't have breakfast today?
No.
Why are you starving?
Okay, fine. You can have something after the session is over. But for now, please be with me. Okay. So active learning is a paradigm wherein you know the gamification aspects can be integrated with tutoring. Okay. So I'll show again a small demo again functions in C but this time you know there is some gamification in it. So you can see that there are some quests and there are some reward points coming you know with each of the conversation that is happening between the tutor and the student. So we can get started with the next one. Yes.
Okay, welcome adventurer to the kingdom of sea. Your quest, master the power of functions. But first, a monster blocks the path. The repetition monster. Ready?
>> Born ready. Let's do this.
The monster strikes. You must square five numbers. The naive hero writes this.
If the spell ever changes, every line must be rewritten by hand. What is the monster's weakness?
>> It's the copy paste. The same logic over and over. If we wrote it just once and reused it, the monster's got nothing.
>> Direct hit. The repetition monster staggers.
That reusable named block of code you described, that is a function. Level one cleared, hero.
The path opens to the spell box. A function is like an enchanted box. You feed something in, it works its magic, something comes out. For a box that adds two numbers, what goes in?
>> Two numbers go in. And what comes out is there's some obviously >> the spellbox glows with approval. Inputs in result out. You understand its magic.
Inside the library temple rest spells forged by the ancients.
You have wielded one already countless times.
Name the spell that prints to the screen.
>> That's printer. I use it all the time.
Wait, print F is one of these ready-made functions, too.
>> Now you see it. Print F and scan F are library functions forged once by the ancients. Free for every hero to call.
Temp cleared.
Okay, so I'm stopping it there. But I hope you got the crux of what gamification is, how you can mix tutoring with gamification. So most of the times we see that gamification this active learning is not very applicable to higher ed but it is applicable to the K12 education especially the primary level education. Recently I had the privilege to visit some government schools in Hana. So I live in Pilani which is Rajasthan but Hana is very close to us. So I visited and we saw that there is something called samp foundation who is you know inducting a lot of videos for students uh you know for students of class 2 3 4 and and teachers are using those videos to teach to students and they the primary you know paradigm behind those videos was this active learning and gamification.
So small kids you know you show you you show this kind of uh uh you know you have this kind of learning wherein there is gamification induced as part of their learning process their attention you know increases their absorption increases their learning increases. Okay in higher education yes there is a challenge in higher education this may not be directly be applicable but this is one of the tutoring paradigms that is available. Okay. Anyone has any question here? So far whatever we have discussed, do you have any question here?
Yes. No question.
Yes sir. Please.
That's just to distinguish that does distinguish between teacher and student.
That's also no questions from students. So no questions can mean only two things. One is you understood everything. Other is you not you didn't understand anything.
Which one is true? First one is true.
I'm glad. Thank you.
Okay. So let's move on to the third tutoring paradigm which is called learn by teaching.
So I'm sure you have experienced this.
Again the one night before the test thing there is one good student teaching to all of the students. Is there anyone like that here? Please raise your hand.
I want to know your experience. Anyone here? An intelligent one night tutor.
Come on. At least one guy.
Okay. Some girl raised a hand there.
Are you in or not?
Okay. One one person. So this girl.
Okay. So can you please stand up? I just want to distinguish.
No, I didn't understand who the student was. So, which one among you?
No. So, who is the student then? You don't have intelligent tutors there at the back. Are you the intelligent tutor? One night tutor.
Okay, I'm not able to hear you.
Come on, guys. I think see okay I have this let me gif if I for you I have this swag kit if you volunteer to give me an answer then I'll give the swag kit no okay no problem all right okay let's Let's come back.
Let's come back. So that one person who is teaching to four other students, do you think that's that guy is the topper?
No. No.
Yes. No. Okay. Next question. Do you think that that person understands the subject better than the others? Yes.
That that's a yes. So the learn by teaching paradigm is based on that wherein the role of the tutor and the teacher and the student reverses teacher starts acting like a naive student as if he doesn't know anything and he lets the student explain the subject to the teacher okay I am the teacher there is a student I behave like a student and the student behaves like a teacher and student explains me the concept to me and in that way and I'll ask some follow-up questions of And in that way the student will become better in the subject. Can you relate to this? Right.
So that's the learn by teaching paradigm. Okay. So I don't have a specific demo for that. But I will be showing the demo of one of the tool which my research team has built. It's a very simple prototype. Nothing fancy in it. Very simple prototype has been built which uses the learn by teaching paradigm. Okay. I'll show that a little later. Okay. But before that if I have to study all three of them together there is one common thing in all three of them. Can you mention what? Can you tell me what that common thing is? Can anyone sorry one one at a time please?
Learning. Okay. Learning is there. What else is there? Understanding. Okay. But there is one common uh Okay. Student response is important. But before student who is talking there is a teacher and the teacher is talking to the student one to one right so underlying principle in all these three paradigms is onetoone conversations okay onetoone conversations are needed and in that way the t the teaching can be effective the understanding can be better yes that was my question I I already asked has anyone even participated in such two paradigms but none of you were willing to voluntarily accept that you have done this I'm giving a chance once again if you please tell me and I want to know your experience of it and then I'll give you a spatip okay yes please the girl at the back please tell your experience okay you can come forward maybe you can pass the mic Hello. Check.
>> Okay, guys. Let's listen to her. Please calm down. Let's listen to her.
>> Good morning, sir.
>> Good morning.
>> This is Bumika.
>> Hello, Bumika. Uh so my experience is like we will have a pure enjoyable learning like uh we will be uh saying some pneumonics for that.
>> Okay.
>> For example like uh if there is a syntax we will be saying creating some stories exciting stories which is related to our personal experiences and we will make it fun like a fun and in that uh in such a way we will discuss among ourselves. we will ask at finalize uh at at the end we will be asking like what are the core concepts actually what are the important key details for that and we will discuss later >> thank you thank you so much a round of applause for her please okay so let's get back guys I have some more questions for you so you see this can we scale these tutoring systems to all the students of India Can we do it?
So for that we will need how many tutors? Number of tutors equal to number of students or at least proportionate to number of students.
Is this a reality?
Can we have the number of tutors equal to number of students?
It can never be a reality. Right? So not every I mean how can we fix this then?
Who said this? Please give him a start, please.
Yes, you can replace the tutor with an AI agent.
Okay, you can replace the tutor with an AI agent. And this gives rise to what is known as intelligent tutoring.
You'll have to calm down, guys. Guys, you'll have to calm down, please. I'm also a professor. Right. Please calm down. Yeah.
Okay. So when we replace the tutor in all three learning paradigms, when we replace the tutor with an AI agent, that gives rise to these intelligent tutoring systems. Okay. So what is an AI agent?
So an AI agent is a software program that autonomously perceives its environment, reasons, plans, and takes actions to achieve some specific goals that are set by humans.
And unlike chat bots that simply answer questions, agents use LLMs to break down complex tasks into subtasks and utilize external tools like browsers or APIs to execute them. So these are you know the formal way of defining an AI agent. But if I have to explain in a simple terms, how many of you use the use chat GPT? I mean that's a bad question today, right?
Everybody uses please don't use during the exams. Thank you.
I mean you can use it before the exams to study but don't use during the exam.
Don't sneak in your phone and use it in the exam. That's not the fair thing to do. Yes. So when you're using chat GPT there is a interface which is a chatbot and behind the chatbot there is a brain sitting. Do you know who that brain is?
That's the LLM. And which LLM is behind chat?
The name of the LLM is open is the company which made that LLM. The name of the LM is GPT4, GPT4, GPT5, whatever. So these are the LLM. These are the brains that are sitting behind the chatbot.
Now can I take that brain away from the chatbot? Okay, I'll separate the brain and the chatbot and I will create my own application. All of you are learning fullstack application development. Yes.
So I'll make one small fullstack application with GPT as the brain. Can I do that? And I can make use of something called APIs. You studied what APIs are?
Okay. We can make use of APIs to connect to that brain which is sitting remotely somewhere. But our application will be able to connect to that brain and that and in the fullstack code that we write with the brain as LLM in that code we can induce some extra logic with which you know you can give it a context you can give you can give the you can give that agent that's called an agent by the way you wrap an LLM with your own custom fullstack application that's called an agent and you can actually give it some features you can you can give it the ability to reason you can give it the ability to plan things, you can give the ability to take decisions and all that's an AI agent. Okay. So now all tutoring systems you can use agents AI agents be it be a single agent or be it be multiple agents come you know working in synchrony. So that's called a multi- aentic system. So you can have AI agents and those AI agents can actually replace two things. Okay. Then you know do you think this has the potential to solve the problem of equitable education in India?
Right? By equitable I mean quality education for everybody. Right? Not everybody can come and sit in this room and study. Not everybody can study in KU. Not everybody can study in bits.
There are so many underprivileged students. So AI has the potential to you know uh to make the education equitable and it has the potential to reduce the social divide with all of this and that's where I have been working on.
Equitable education is one of my research areas. That's where I have been working on. I build we build systems and we deploy them. We do studies. We do they're known as human computer interaction studies and then we see how it's benefiting and all. Okay. So few examples of agents. So there can be a personal assistant, there can be you know customer support agent and so on.
Okay. I have one more question. Can you think of an AI agent that you have come across? Please raise your hand. Please don't say it. Just raise your hand.
We'll come to you and you can answer.
Have you come across any AI agent in past few years? Yes, please.
Okay.
Okay.
So, it basically does the full market research uh on the current trends and all. It extracts the hooks and all and based on that it generates the new Instagram scripts.
>> So uh this is like full content pipeline which will do the whole market analysis, script generation, video generation. So it's like I have created my own system.
Yeah, >> that's wonderful. That's wonderful.
Thank you. Thank you for sharing that.
>> Thank you so much.
>> And I'm happy to see students exploring those things. I mean that's how you learn. You should explore those avenues and that's how you learn actually. Yes.
Any other hypothetical agent you have seen before?
Anyone has an answer?
Any any real agent or hypothetical agent which okay he has built an agent. I'm saying some existing agent we have you have seen you have used you have heard about anything cla you can say that's an LLM again so that's a chatbot I'm talking about an agent which is wrapped around an LM or or so the when LLMs were not there then you know a agent still existed it's a different way of doing things lama oh lama I think that's an element.
It's not able to hear you.
I'm still not able to hear you.
>> NAN, >> what is that?
>> It's like an uh AI agent system.
>> Okay.
>> Which uh allows us to build multiple AI automation systems and all.
>> Okay. Okay. Thank you. So, okay. You have something to say at the back there.
Okay, let the mic come to you please.
>> Sir, I came across the open claw. It is a one a aentic framework.
>> Okay. Open claw. Yes, I heard of this.
in which we use ALM API which as a brain and uh it actually does work.
>> Okay, good. Thank you.
Yes.
>> Okay, you can please talk in the mic please. Everyone can hear me.
>> I have seen an AI agent developed by IIT Roper. Please calm down guys. Please listen to him.
>> Uh an AI agent developed by IAT roer which actually uh takes the agriculture development and etc. And I have also worked over there for small uh research and things.
>> Okay. Good.
>> By sudar shans at IO.
>> Okay. Now let me change the question slightly. Now I want you to find one agent from me for me from Hollywood.
Anyone?
There is one very famous AI agent from Jarvis. There you go. I was expecting that answer Iron Man.
So 15 20 years back there was an hypothetical very powerful agent called Jarvis in the Iron Man series.
Okay. And I'm telling you even today we have not reached Jarvis level of intelligence. Okay? We are almost there.
I published a paper two years back. The title of the paper is it's not like Jarvis but we are almost there. That's the title of the paper.
Okay. So yes, we are going closer to closer and closer to Jarvis day by day.
We're almost there. But even today that hypothetical Jarvis was very powerful.
Very very powerful.
Okay fine. Let's continue.
Yeah. the advantages of intelligent tutoring systems. So I had mentioned one of them is you know you can we can have equitable education wherein quality education can reach larger masses and it builds an equitable society. The social gap can be bridged and most importantly you know with these intelligent ting systems they can adapt to learner space needs and understanding.
you know a lot of customization can be done learning can become very personalized when we you know we properly you know use these AI agents for tutoring learning can become very very personalized that is one of the biggest advantages of having intelligent ting systems okay so I any question here so far before I move on to this is the multi- aent system which we have built we'll not go deep into the tech part of it but I'll just explain what it is any questions you have Yes.
>> Hello. Good morning, sir.
>> Good morning.
>> I wanted to ask how can you make uh it equitable with AI because uh people who are with lower background, they can't even they don't even know AI. They can't even use AI properly.
>> Yes, that's a good question. So, that's where the outreach programs have to come in. Funding has to come in. And uh so you do you understand the difference between a frontier model and a small LLM?
So small LM are free you if you can uh fine-tune those small LMS then all those rural population can get quality tutoring at free cost because you don't have to spend on those APIs right that is one way with which you can do but larger part is you know the outreach programs which governments and NOS's have to actually take take up and uh take it to the larger masses but once taken it can be you know very good the potential potential has it has a immense potential.
>> Thank you.
>> Thank you.
Okay. So, let's get start. Let's uh go to the next part. So, I'll I'll take 10 15 more minutes. Is that okay? Or am I boring you too much?
Yes or no? Wow. Thank you. You have something to say? Yes.
sir. You said that LML >> LLMs are small LM LM.
>> It's okay. All of your friends here.
It's okay.
>> Those are free. No sir, you said that those are free but they they give errors also. No sir. Sometimes >> that's why you need to do finetuning on them. No sir, >> sometimes we use the free chargy. No sir.
>> Yes chat GPT you can't build a custom agent.
>> Custom agent would need you to build some write some code and for that you'll need to use the GPT40 API but that is not free. But if you have a small model and you have fine-tuned it sufficiently with quantization techniques and uh low cost with quantization techniques you can do low cost inference due to which you can actually do everything on your own computer or the computer that's provided by the school today.
>> Thank you.
>> Okay. So okay I'm very happy to see good participation. Okay. We'll take questions later. Let me finish this. So happy to see the participation. So I'll take 10 15 more minutes and actually give you a brief idea of what this teachtolearn system that we have built is. Okay. So yes, please don't talk amongst yourself.
>> Please don't talk amongst yourself.
Right? Let's respect professor Jagat is standing there and talking to you. When he's talking, please don't talk amongst yourself. Thank you.
>> Okay. So we saw three tutoring paradigms. The last one was learn by teaching and we did not show any demo for that. But we have built teachtolearn which is a multi-agent tutoring tool that uses learn by teaching paradigm.
Okay. So I'll show you I mean I'll run through what teach is and I'll give you a small demo of how it works a very miniature demo not major one. So yes. So what teach tolearn is so it's a multi- aentic framework wherein uh you know you can use this system in order to revise your lessons before the exam that is one of the use cases. So this does not replace classroom teaching but this you know augments classroom teaching by uh you know by providing this this you by providing this tutoring aspect. So what happens in this tool is you actually select a topic which you want to learn today or I would say which you want to revise today. Then the AI agent will generate some questions for you and the AI agent will keep giving you some questions and there will be an interaction between you and the AI agent. You explain concepts to the AI agent and AI agent will ask you more and more questions. There will be a conversation between you and the AI agent. Okay, you teach to the AI agent basically and then towards the end the whole conversation is assessed and some feedback is also given. That's the simple workflow of teach tole learn. I do have a slide that actually shows the architecture of this but I don't think you will be able to relate to it. So I've hidden it. Since you are first year students you might not be able to relate to that architecture. So I've hidden that slide. I'm not going to great details but we we used around five agents to build this model. Okay. We can build it with three, we can build it with four, but we decided to use five agents and it works with five with that way actually.
Yeah. So now I said that you are teaching to a tutor. Say in learn by teaching what happens? Student teaches to the tutor. Tutor although is very knowledgeable about the subject, he will still behave like a student, right?
only then he will be able to listen to you and give you ask follow-up questions like any other student. Okay. But here I'm saying I'm replacing tutor with AI system. If I'm replacing tutor with AI system, how do I induct the knowledge to the to the agent? Actually student bot or agent is the is the t is actually tutor actually who is tutor who is acting like a student actually. How do we induct a particular subject? So for example in the videos that we have seen we were discussing about functions in C right now how do I teach AI what functions in C are what do you have what do you say see I can't blindly bring in an AI system and just put it replace it in place of teacher I should also ensure that whatever the teacher actual physical human teacher knows AI also knows right I have to ensure that only then the AI agent can actually behave as a tutor in any paradigm you pick up how do I do that? So for that there are two technologies one is called rag which is retrieval augmented generation and other is the finetuning of a smaller element.
So if you use these two technologies you will be able to induct the contextual knowledge to the AI agent. Okay how many of you know the book HIVA?
Very good. So what we did I mean this is the side thing. What we did is we scraped the HCHMA book and made an intelligent bot which can actually behave like a physics tutor. It can even generate those equations and those mechanics and all of that.
How did we do? We used rag wherein we scraped the textbook and we inducted that knowledge to the LM through this method.
Okay. So rag is very powerful and fine-tuning is also very powerful. These are very powerful techniques with which you can actually induct some specific knowledge of your choice to the tutor and then you can ask him to behave like a tutor in any paradigm you want or socritic or lbt or whatever.
Okay, that's the idea. So you can save these words rag and fine-tuning. You don't need to understand what these are today. It's okay for first years it's okay. Even if you don't understand it's all right. But as eventually you progress with more AI related courses, these will come and then you'll then you'll be going to appreciate the way they work and the way they support you know these systems.
Okay. So last demo just only one small demo I'm giving of the application which we have built in our research team. So it is like six seven minutes and this would be one of I mean after this there's almost nothing. So please pay attention very small demo. This is I have not I have purposefully kept it small so that you know it finishes quickly finishes on time. So this is the bot that we have built uh it's called teach tolearn and uh this works at the paradigm in the paradigm of learn by teaching by looking at this interface.
Can someone guess what this is?
Which framework was used to build this?
Can someone guess?
Yes.
Louder.
Sorry.
No, no. I'm saying the interface that you're seeing, the screen, the chatbot, the black color. This interface you're seeing, do you see this in any of the frameworks? You since you've done fullstack, you would have seen something.
You heard chain lit.
The word chain lit. You heard streamlit, chain lit, right? This is chain lit actually. So these are used to chain lit is used to build, you know, these kind of chat bots. Okay. at at the back end you can you have uh lang chain at the front end you have chain lit that these are the technologies which we use to build a rag based application okay so yes let's get started with the demo I'll not get into too much of details please listen I mean it's a very small very simple it's a prototype only so I'm going to give you a small demo of this learning >> okay I think I will need more volume can someone help me with the volume The volume in this video is very less.
Can someone raise the volume here?
Max, this is Max.
Hello everyone. So I'm going to give you a small demo of this teach to learn tool. So as explained teach to learn is a tool that is built on the paradigm of learn teaching wherein as a learner please listen I will be you know enhancing my understanding of the topics. So this bot this application has been trained on four subjects. So we have computer programming we have data structures and algorithms.
>> Please maintain silence and please listen. Thank you.
I'll give you a brief demo about uh let's say I'm revising some particular topic of computer programming. So let me select computer programming.
Yes. So here are the topics in computer programming which I have to select. So I'll select data types, variables, constants and operators.
So it's generating questions. So the student bot will ask some questions to me. So here is the first question that it has asked. What are the different data types in C and why do we use them?
So let me give the answer to the AI part.
So C has multiple data types.
They include int, float, double, short, cap, long, long, long, long, double, etc. Okay, int is okay, let me go to the next line.
Int is used to store integer values. Float and double are used to store floating point or real values.
Car is used to store a character.
uh car array can be used to store strings.
Long and long long are data types used to store large integer values and short is used to store smaller integer values. So based on my understanding of this question, the answer to this question. So I have typed a reply. Let's see what how the bot behaves.
So I send this message.
So now the bot is act acting like a novice student. So it will digest this answer and it'll ask some follow-up question. So yes, it said hold on something just popped into my mind. So this is a follow-up question which the B is asking. How do variables differ from constants in C programming? So that means I have explained different data types. Now it is asking me about constants.
So let me give the answer. So variables are placeholders to store data that is mutable.
So this is the key difference. So let's okay can you provide an example when to use static storage in Cclass now it's asking me questions related to you know the uh the storage classes the static storage class so static storage class can be used to create variables that will reside in data segment of the memory.
Meaning the lifetime is global. Static variables reside in the memory as long as the program is executed.
They are created once only.
Static variables can be accessed in the specific block they have been created.
However, even when even when the block gets terminated, the static variables still reside in the memory.
This is unlike a global variable or I would say local or or an auto variable which dies out when the block finishes its execution.
Okay, this is the answer that I have given. Let's see what is the next question.
So yes kind of it is somewhat satisfied maybe it's giving me final remarks. So it is giving me strengths improvements and summary. So it says that the teacher consistently provided clearer explanations as evidenced by teachers clarity scores of 7.9 and 9. So these are the scores for the three interactions that have happened 7.9 that means the bot is able to sense that you know I I have better understanding I have good understanding of the topic which I just talked to the bot this clarity helps maintain a high level of student understanding and engagement.
So this is a very active research happening. In fact, I'm part of one global research consortium where in seven universities across the globe are studying on the impact AI has created on the education sector specifically skills, cognitive layers and human- centered uh things like you know trust and all of that. So we are seeing how we are trying to understand how these things are affecting them and we are also trying to come up with a policy with which the adverse effects can be mitigated. So that's also an active research. I'm part of that also. So I didn't present those results here because you will not be able to relate because you're first years that those things will be you know more appreciated by you know faculty and admin staff but may not be by students. So therefore I didn't bring in that but that's a very you know active area of research where we are trying to understand how it has impacted and how those effects can be mitigated. See air has come so you can't go back right any adverse effect is there we have to understand how to manage it and that's where our study comes into >> my personal >> first question about environment thing so you are saying since you're using five agents isn't it increasing the carbon emissions my answer is no because five agents were being used doesn't mean that five different models are being you know at a time uh you know being used or anything. So the number of inference calls that I'm making or number of API calls that I'm making pretty much would remain the same even if I'm using one agent whether I'm using one agent or whe whether I'm using two agents the number of API calls that I make to simulate this that would remain the same. So therefore this carbon emissions is directly proportionate to the agent the I won't say it is directly proportionate but to vaguely tell you it is directly proportionate to the number of API calls I'm making but it's not actually that okay >> my stand is just to say that we need some time more time to get the research of carbon emissions and then we need to go into the scaling the >> I would agree with you that should be I mean I'm sure that is another active research so coming to carbon emissions even this blockchain technology have done a lot you know just for mining one bitcoin you know how much of coot is released lot of comput power is in that way any technology that's coming this issue is there even electric vehicles are set to be clean but as on today I'm not sure but okay I can I can tell with some confidence that the carbon emissions net carbon emissions after after bringing EVs has not gone down Because India is still dependent on power with carbon right so you burn carbon there how much is the efficiency how much of energy that you're burning is actually getting into converted into electricity there is loss that loss is more than the saving that you're making right so there is lot of research going on and lot lot more to go okay thank you for the question please Yes, >> good morning sir.
>> Good morning >> sir. I wanted to ask about how those five agents are collaborating with each other and if any collision is happening between them how you are resolving inside it.
So I said I have a slide which I have hidden from you right? So there is another slide.
So these are the five agents that are there in the system. One is generating questions, one is a student agent, one is an evaluate agent, then the scorer summary. All of these are different agents. Okay.
>> Sir, but if any collision happens between them, then how we are handling >> framework the architecture is designed in a way that there is no collision. The interactions between those agents follow this architecture. So there is no collision between them. They are streamlined.
Okay.
We are giving them autonomy but at the same time we are ensuring they're able to coordinate. That's the idea.
>> So can't we so can't you only uh fine-tune an another LLM to if to keep keep a check of these collisions or >> you can but that will not happen.
Collision will not happen.
>> Okay. But the answer in principle is yes. For example, there is something called RLHF. You heard of this word reinforcement learning human factor.
>> So that that that is like RLHF is like you finetune another small LM in order to filter out the unethical things which GPT will produce right small like another small LLM is there. So you can do that actually okay don't worry guys this figure you will not understand. So I don't want you to understand therefore I but I want to show this thing. Okay. Yes sir.
Uh not so far sir but that's an important thing which you raised and that's there in our research agenda too but as of now we are not.
So this is this QR take to my LinkedIn page. So if you want to connect with me this is the place and uh my email id is also there. If you want to write to me for anything most welcome. Okay. So I think I'm already running over time by 8 minutes. Okay, one last question. Yes.
>> Good morning, sir.
>> Good morning.
>> Uh just now you have shown the last demo right?
Am I clear?
>> Yes. Yes.
>> Uh just now you show the last demo right? For example, in our day-to-day exams, we are most of the uh who the teach in meanwhile exams, right? So how in the same case the if the uh person student is using the charging how can that be corrected and how can we genuine the scores >> okay so first of all the demo I have shown is an agent which behaves like a student and you are teaching to it when you are using chat GPT you raise a question and you get the answer that's not the purpose of this that let's say that is stage one you are learning from chat Chad gubity stage two is your teaching to that chad jubility this is stage two teaching to chad jubility to strengthen your understanding again it's based on the principle that when you when you teach to someone you become you become better at that subject anyone who has taught to their peers would experience this I think some of them actually told that they have experienced this yes this is true okay it's based on that paradigm >> okay sir I'm saying that the you have scored you have displayed the scores right >> yes >> the scores can't be genuinely Even if we use the charg for answering right >> yes I mean now it's up to you whether you really want to learn or you really want to cheat not cheatingness is not in the exams but you know in a in a game sense okay yeah it's up to us so that requires training that requires education okay thanks okay so few more questions at the back and then we'll stop there two more questions and that's all because I'm already running out of time there are other events that are being planned Yes, at the back.
>> Good morning, sir. I have two questions for you.
>> Okay.
>> Uh first one regarding your agent.
>> How have you >> how have you modeled the agent uh to take new inputs uh regarding newer publications and all the uh >> Okay. So, it's not automatic. We'll have to train.
>> So, we have to train. Anything new we'll have to train. Yes, >> you have to input every single publication or anything.
>> Yes. Anything you want to become you want AI to become expert on you have to train it separately either through rag or through LM or through fine tuning.
>> So my second question is a little different. How can a user um use the tokens efficiently?
Because um if you use claude, if I ask one question, it will generate a log and the limit will be off with just one question without even fulfilling my build.
>> And so my question is how can the user efficiently use his tokens?
>> Okay, so before that I would want to tell that uh that's why we are working on creating fine-tuned models. If you have fine tuned models, you don't need tokens. You don't need anything. The fine tuned model is there existing on your computer. You you call you make any number of API calls as you like. But these frontier models, they charge you for number of tokens. Right? So first of uh so if you replace frontier models with a fine-tuned smaller models, this problem is completely solved. When it comes to token management, I do not know the specifics, but some models do support lesser costs when you, you know, when you do batch processing, right?
Generally, they come down to 50% or something like that. But it also depends on your application whether batch processing is even possible or not. For example, in this it was not possible.
Okay? So therefore, you know, if we have to use frontier models, then there has to be some funding that's coming from the government or somebody in order to, you know, keep those systems alive and you guys can use it. But one good thing is the money is not so much when compared to you know spending on training somebody else.
Yes please >> sir as a is a developed as we can take the practical example like the cars which have been automated cars.
>> I'm not able to hear you clearly little louder >> sir. Huh sir if we take the practical example of the automated cars like Tesla company but it is not suitable for the Indian road. It is making some of the accident while using the AI.
>> Yes like technology is making the accidents in the Tesla companies and so many are facing so many difficulties while using the AI. So many are facing the difficulties while using the AI in the car technology or any automated technology.
>> So what's the question?
>> My question is AI is missing some of the mistakes like accidents like this.
>> Yes. So that's why you know AI is there but it's not like it's not like I I wouldn't say okay you understand the term artificial general intelligence. So that is a much more advanced thing than AI. So we are not yet there towards AGI.
AGI is very much much much more advanced and we are maybe at 10% level of AGI. So as we progress closer and closer to AGI probably these things would reduce but we'll have to see wait and see how it behaves. But when it comes to suitability of these automated driving systems for the Indian roads one thing you should know that all these systems are mechanical right.
>> Yes sir. These systems are mechanically they will behave the way you train them.
The the level of customized training that is required on Indian roads is very very high and so far we are not able to you know there are some companies who are working but so far we are not able to achieve any production ready version.
Yes sir. Okay. But if you ask me whether it is possible, I would say with some hitch that it may be because you know if you understand how model training happens and all then you know that hitch will be there. We'll have to wait and see. Okay. Okay. Fine.
Thank you very much. You have been lovely audience. Thank you for uh keeping it interactive and uh yes this is my email id. I am happy to answer any emails uh that you have. Okay. Thank you so much. Over to All right students, so that concludes the first session for the day. We will take exactly 15 minutes break now. All right. Please come back to the hall by 11:30. Sharp 11:30. We will start with the second session which is a hands-on workshop.
Exactly 15 minutes, please. If you enjoyed this session, you are really going to enjoy the second session where we'll actually build an agent.
All of you, can you please wait for just one minute? Let's take a picture.
Let's take a picture. All of you, if you can please sit down.
Can I request all the students to please sit down for just one minute. We'll take a group picture and thank you very much for your participation for all of your questions.
I think it was your participation that made the session really really interesting. So thank you very much for that.
Perfect.
Students, please take a quick break and please come back by 11:30 sharp.
We need to be on schedule if you have to maximize what we want to achieve from the workshop. Please take a quick break.
We have 5 minutes from 11:30.
So if some of you can maybe start rounding up the other students please.
It will take us 1 hour to finish the first part of the workshop and then after lunch we will continue. So 1:30 again we will start and then we actually have something now in this workshop.
Right? All of you will be opening your laptops. You'll be building something.
If you have any common WhatsApp groups or anything if you can please just message there for the rest of the students to come back to the seminar hall within 5 minutes that will be very helpful. Thank you.
Heat.
Heat.
Heat.
Heat.
Heat.
Heat.
Follow Heat. Heat.
Heat. Heat.
Heat.
Heat.
Heat.
Heat.
Heat.
Heat.
Students, please settle down quickly.
We'll start in a minute.
Heat.
Heat.
Heat.
Heat.
Heat.
the market and then we go into some product thinking framework where we give you some tips on how do you solve a problem right how do you pick a right problem to solve for and then we go into the building workshop where we'll use claude to understand promptic techniques better and then we'll build prototypes using lovable right like Jagat uh professor Jagat said he showed you a pro prototype today we will build the prototype right based on a problem statement you guys choose and Then we'll just uh cover some logistics on some of the course we have created for you. But for now I'll just hand it over to Kushboom.
Thank you Prane. Can I please request all the students to settle down quickly please take your seats and please don't talk amongst yourselves because if you do that you'll miss what I'm saying and then it will start making less and less and less sense as we go through the next slides.
All right. So, two scenarios here in front of you, right? Let's make this interactive. Don't think of me as some faculty. I'm not a faculty member. I am also learning. I work with Corsera.
Think of me as a senior student. Let's make this interactive. Please answer.
Right? Two candidates. There is candidate A. He's a top coder. He or she is a top coder. Expert in their field.
Right? A student throughout 10 CGPA.
perfect to the tea. That's candidate A.
I have another candidate B who might not be, you know, Mr. Perfect or Miss Perfect, but they use AI tools to build demos. They talk to users. They know something about everything.
Can you tell me by a show of hands, how many of you think candidate A wins in general in their career, in their jobs?
How many of you think candidate A is the better choice?
How many of you think candidate B is the better choice?
How many of you think none of them is a good choice?
Okay. How many of you think it depends on the situation?
Yes, that's the right answer, right? So it depends on the context.
If a recruiter wants to hire an engineer to build a large language model, they need someone like candidate A. They need a perfectionist. They need somebody who really understands their concepts perfectly.
But if you want to build a startup, candidate A might not be the perfect fit because they are a great coder, but if they are building something and they don't know how to talk to their users, how to gather feedback, how to understand operations, you know, a little bit about everything, they might not be the best fit for a startup environment. So it depends on the context.
Yes, students, once again, please don't talk amongst yourselves.
All right. So this is a Hindi phrase.
Some of you will relate to it. Those of you who know Hindi, can somebody just tell me what is this Sharma Ja beta?
What does this meme mean?
Perfectionist. Someone who is an A student in everything in life. A in games, A in sports, A in marks, A in everything. Right? This is what our parents used to tell us. Be like that student. So this is that student. The fact is that the ideal career trajectory is being challenged and this is no longer the norm. So you don't have to be this student. There are other ways of excelling in your career. Right? Be the best coder is not the only answer anymore. Traditional belief has been that deep skill means you're successful.
If you have a moderate skills but you have other multiddisciplinary skills, you may not necessarily succeed. Right?
That's been the traditional belief.
reality right now is that AI is compressing skill advantage, right? So, everyone has access to the same knowledge and tools and the result is that being only good at coding is no longer the only mode.
There are multiple ways to success.
You can be candidate B and be a very successful team member for a startup that is building something from the ground up.
Right? You're not competing with just your classmates anymore. You're competing with AI and everyone who knows AI.
Now if you were to look at this skills framework right this is what we call the future ready skills framework. What exactly is this right? So if I look at the skill buckets these are the five skill buckets that employers today look at the first one being technical. So obviously you have to have some solid foundational knowledge. You might not be an expert coder but you have to have your foundations rock solid. You have to have AI literacy and of course engineering hygiene which comes with your coursework that you're doing as a part of your university degree. Right?
Why it matters in the AI era era is because AI tools speed up output but only if you can specify, test and integrate. So AI helps you but it's not going to do your job for you ultimately.
Right? AI is a tool that you have at your disposal.
The second skill is cognitive. So problem framing, how to synthesize a problem, how to articulate a problem effectively, systems thinking, critical evaluation.
These are all things that you need to be good at.
Why does it matter? Once again because LLM output is cheap but correct decisions are not.
Ultimately the decision on a particular problem has to be taken by you the human that is working on the problem. LLM helps you with that process but it cannot take that decision for you. And in order to take the right decision what do you need? You need cognitive skills.
Right? Third which is social skills. So communication, teamwork, stakeholder translation, right? If you're building a project and you need funding for your project, let's say in your third year you're building something amazing, you need funding from the university. How are you going to actually build a project plan? How are you going to build a business case out of it? How are you going to communicate to your team of four or five students who are working on that problem? How are you going to appro approach your head of department or your deans or your um you know your senior leadership and ask for that funding? You need to have the right skills to be able to do all of these things. So communication, teamwork, stakeholder translation and why it matters is because generalists win by reducing coordination costs right. The fourth career habits of course, right? So if you're consistent at something, if you are good at following routines, it obviously means success.
And this applies to everything, right?
If you want to be good at a particular topic, at a particular skill, you have to do it repeatedly and consistently.
If you want to learn programming, you have to spend some time on it every week. If you want to build muscles, you have to go to the gym a few times every week consistently. It will not happen overnight. Same goes for anything in life, right? So good career habits and why it matters is because career outcomes often hinge on initiative and signaling not just on marks right and finally high agency. So what does high agency mean?
Um what do you want to take this agency?
>> So high agency basically means that your professors are not asking you to do things but you yourself are trying to find problems solving them. the curiosity and basically being accountable, right? Like that's what you have to signal to your potential employers, your faculty that hey this person is reliable because when you go out in the market technically they are hiring you to do some job but they should get that confidence that this person can do that you should be reliable and how do you build that by building high agency right and it also comes from permissionless building and that's what we are here for today right we want you to think about hey how can I solve for a problem today I'm facing and build something and show it to my faculty, my potential employers, right?
That's going to make the difference. And why it matters is because high agency turns breadth into shipped outcomes. You can go in the interview, you can tell hundreds of things that I'm good at this, I can do this. But if you show something that is worth more than all of those things combined. So that's why agency is important.
>> Yeah. Product yours, right?
>> Okay. So you guys are anyway here for the next 45 minutes make use of this time right guys I need your attention so what we'll do now is actually build something hands-on but before that I want to like ask a question in terms of like how do you guys identify a problem worth solving anyone right is there a mental model in your mind that okay this is a problem should I really solve it or not what do you generally think when you see a problem it can be any problem you face in a day-to-day uh like student life or anything outside your >> so you can please raise your hand just think of it whatever comes to mind right how do you identify a problem that yes this is really a problem I want to solve how does that process work for you >> can you just stand up >> you can just stand up and speak yeah >> personal experience all right >> so in personal experience like when you see a problem what is the first thing you think about available solution.
>> What if there is no solution?
>> And how do you decide if bridging that gap is worth it or not?
>> So basically what you're saying is if it will be useful for more people, right?
That's good. Anyone else?
Uh, sir, there's somebody in the back.
If we can get the mic sir, first I will understand the problem by going into the system and understand the problem at the root level.
So basically what you're saying is before even jumping into the solution I will try to understand the problem at depth. Yes sir. At the root level >> and uh you have understood the problem.
How would you approach it now? Like you know that okay this is a problem that exists and it's a uh problem worth solving. But how do you decide that?
>> Sir if uh if my solution solves uh any kind of problem like uh if it saves some time or if it generates some money and uh then it will be good solutions.
>> That's good. Basically what he's saying is I should see some value out of this.
Either it is saving time for someone or it is saving money for someone right so that's what we are going to deep dive today a bit into product thinking right because this kind of thinking is what will differentiate you from lacks of other students right companies want product thinking there's when you go work in corporate or anywhere else for that matter it's not like you're going to have a laundry list of 10 projects and you have to finish that that's not how it works you need to identify problems that are worth solving for and also give solutions to them. So building that muzzle again this is something which doesn't come in a day right the more you practice the more you think about solutions for example you're using Instagram right why did someone create Instagram they saw a gap right they they were seeing something about making connections making friends and that's how Facebook started and from there Instagram came so you have to build that muzzle so instead of asking what should I build you should start thinking about what problem is worth solving doing right so one scenario is okay I want to build an app I want to make use of AI I want to learn this technology this is a tool first thinking basically you're jumping directly into the solution but instead right what good builders right think about and even for that matter good software engineers think about is this is the user is struggling with a problem and they start their thinking like that right this process is broken this takes too much money this takes too much time.
So that is the product thinking. Code is only a tool but problems is what creates the value.
So I want to give you a simple framework right. How do you identify if a problem is worth solving for? Number one pain intensity. Is this actually painful or the solution or this particular problem is a nice to have. Right? The more people have this pain or the more people feel this problem, that is where you're talking about the frequency. Let's say you're thinking about uh project submission as a problem, right? There's project you have to uh kind of write a record or something end of your four years. That is a one-time problem in four years. The frequency is not that high. But if you're talking about something to solve for your everyday quizzes, right, everyday projects and all of that, that's a very high frequency problem. Now the third one is this is important the willingness to act. So with AI all of us can build stuff all of us can build tools but is someone willing to use it or is someone willing to pay for it? That is the problem worth solving for. Right?
Because you can build hundreds of tools no one is using it. It's all waste. So what you need to think about is a combination of all these three. Is the pain intense? Is it a really painful problem? Second, does it happen often?
And third, is someone willing to act on it? If you find a problem statement like that, then that's a gold mine.
So, let's just kind of uh put this framework into a statement. What you want to do is identify the user. Who is the user that is facing this problem, right? And then what exactly is broken?
You need to be able to articulate that particular problem. And the outcome is what does the success looks like? The final output basically what you want to get to is I want to solve X for Y that does Z. So now what we going to do for the next 5 minutes is we will identify a problem statement right and we will use that problem statement to build something today in the workshop.
So let's start with think about your career or like your next four years within the campus.
Let's think about what are the problems or what are the key milestones you want to achieve within these four years and let's start from there like what do you think you will face when you are uh going into placements or what do you think you will face when you are preparing your resume.
So let's think in that direction and let's outsource some problem statements.
It can be any problem statement guys.
>> So you're all first year students right?
You've already completed two semesters.
Think through the last 12 months when you joined Klu to today, right?
What is the one problem you have faced in these two semesters?
Something which is painful enough for you. Something which repeats again and again and again. And something that you want to solve. Think of that and put it clearly into a statement.
>> You can raise your hand if you want to share. Yeah.
>> Let's let's have this interactive guys because we are going to build what you are asking for in this session itself.
>> Anyone?
>> Don't worry about perfect answers for now. Just just brainstorm your ideas.
Think of any one problem.
And if you have any future vision also that is also fine right it doesn't have to be the problem you're facing this year maybe third year fourth year okay let's I I'll tell you what problem I faced when I was a student right one of the things is preparing for an interview right I used to like uh reach out to my senior alams and try to find some slot with them asking for hey can you re review my resume or can you help me understand how to prepare for this interview? The problem is all thousand students are reaching out to those 200 uh alumni itself, right? You can't get one-on-one mentorship with each senior.
So that was a big problem for me.
Something similar to that any thoughts any problems? Again, it it can be something outside your uh education as well.
You had something time management issues. So that's a good point. But you see the problem with that statement. Time management issue is a very broad topic, right? What exactly in time management issue are you talking about?
So you're saying during the placements you have to balance your studies and uh you also have to prep for placements.
Okay. And why do you think that is happening?
>> Okay. So, you're saying you cannot give time for both of them. But why are you not able to do that? Is it because there are too many courses for you to study for or too many placements for you to prepare for?
>> Okay. So you're saying you have to practice more and more coding and so that you can get better placements. So can you do one thing? Can you try to frame it in this way?
Right? Take take some time. Think about how would you frame it for this. So I want to solve for me right for this problem that helps me do this. Right?
Can you try think about what that statement would be like? Because see the thing is problems can be very generic.
What you want to do is actually pinpoint and say this is exactly what I want to solve for. It can be a very small workflow or it can be a very small problem. But if you cannot pinpoint that this is the problem you cannot solve for it because you have to break it down into components or you have to break it down into smaller flows and then build for it. So think about it right I'll come to you we'll discuss how you can frame that problem statement. Anyone else?
and also type of diversification of what I have like uh let's say I want to apply for an internship for the project management I need a different resume rather than what I apply for a research thing and also if I need to apply for an university and also to a company, I need to differentiate my resume and yeah, things like that.
>> Okay. So, what he's saying is I don't want to send my same resume to different companies or if I'm applying for a research assistant uh role. So, he wants he feels that it's a problem that I have to repeat my resume and create my resume for every company. That's a good problem, right? What else? and clashing my schedules with I mean missing my schedules.
>> I think that sounds like a personal problem but yeah. So what do you mean by missing my schedules?
>> Like I have an exam at 9:00 today and I forgot that I had an exam or something of that sort. Keeping my own schedule.
>> I think they solved something for that already. Alarm.
No, but >> basically what you're saying is you want to build in uh I want a buddy who can remind me, >> right? Automation. Yeah.
>> What else? Anyone else? Yeah. Can you give the mic to him?
>> We need three more problem statements.
Come on guys, more participation.
It can be a very simple problem. One general problem statement that I want to uh uh tell you is that finding out the resources when preparing for any uh government exams or for any exams is the one of the major problem everyone will face.
>> Can you explain a bit more on that? What do you mean by resources? like uh for example let's take the exam UPSC like uh so many people will attempt but we will get a perfect uh guidance for the from the people who clear that but generally we we are not we are not having the contact or personal touch with them. So uh for the people who uh wants to attempt for theuh future attempts they must have to know what the plans and what are the resources they used for that and what are the best uses resources uh in the current generation all so many resources are there but uh some are very useful some are very lagging so what are the best resources they want to they used for and what they want to suggest for their juniors or upcoming one.
>> Got it. So basically what you're saying is you want to solve for a mentorship problem for as UPSC aspirants so that they can get advice directly from people who have actually cracked it. Right? So that you see what I did there. I tried to frame it in such a way so that the problem statement is clear. Yeah. No, that's a good one. Right? What we'll do is in this session we'll try to brainstorm how we can solve for something like that and try to build a prototype. Stay with that. But anyone else?
Yeah, >> please pass the mic.
>> Instead of that, if I focus on building an application or being a full stack developer, it's a waste of time.
>> So that's >> basically what you're saying is that I want to solve for career guidance, right? specific career guidance which is personalized to you so that you can crack that dream job you have right so that's what you're trying to say okay I think we have good number of problem statements let's move into building this right so there are some ideas your classmates gave either you can think about that idea or you can think about any idea you have but how do we start from here right we have a problem statement what AI does really well is acting as a thought partner. Right? All of us use chat GPT, Gemini, Claude on a daily basis. Right?
Now, one of the things we in my work also use on a daily basis is when I identify a problem statement I use LLM as a thought partner, right? I how do I fine-tune this? You said you want to build a career advisor, right? Uh you said you want to build a resume uh prep uh agent or a tool, but how do I reach there? what exactly should that contain?
What exactly should the should the solution be? Right? That's where I use LLM for. Before we go into the next slide, I want to just quickly check with all of you. Are there any techniques or uh what do you call tricks you use when you're prompting the LLM? Right? How do you generally start your uh prompts with an LLM? Do you use anything specific?
Yeah.
First I assign the role then pro I proceed with the task and then with the conclusion like what actually I need with it. That's how I uh give the prompt to the LM.
>> That's a very good point. Can you explain all of us what do you mean by role and why do you do that? So basically when I'm building an application I assign the role that you are a senior uh front end as well as backend team where you are uh building an good application then I'll proceed with the task which I so uh suppose I'm building an resume like as he said that uh or like application of a resume building thing. So I'll suggest him that your task is to build an AI based resume selection system which will uh give our like uh generate a resume based upon our uh like experiences and all. So the final would be like uh what uh things which we need to use which things which things we are going to proceed with these all things I'll go with.
>> Thanks thanks for that I that's one of the things I wanted to cover as well. Uh but >> thank you so much.
>> As we use LLM daily, one thing you'll notice is for your example, you give your resume and ask it in a such a way that I think this is the best resume out there. You tell it, it will confirm with you, right? It'll say, "Yeah, this is the best resume, you'll get all the jobs you want." But that's not we want. We don't want a liar, right? That's where we need to use some of the prompting techniques. So the first one is assign a persona, right? That's what he was talking about.
Tell the LLM to think like a research assistant or to think like a front-end engineer or to think like a career adviser, right?
What that does is because LLM's hallucinate all the time. Have you heard this term hallucination, right? Like in you'll say, right? It it just says anything you want it to say.
So what that does is it will create a mental model and it will steer away from generic chat. So that's why it's always important to like assign a persona. I mean I just tell it like you are a you are the best-in-class let's say front- end engineer with 20 years of experience. That's how I start my prompts with right. The second one I think this is most important. Give it constraints. Don't just give it all the free rent. Put rules in place. Tell it I don't want any generic solution. I want something I can create in the next two hours, right? Or I want something I can create in a week. This is my skill set.
So suggest me something I can do with that set of skill sets. What it does is it won't give it the entire creative freedom but still puts in guard rails.
So it's always important to give constraints. Third is the quantitative formatting basically force the AI to use a scoring logic. Now where this is important is let's say you're building a solution or let's say you are uh writing an essay assignment it gives you 20 points right for an essay but your word count is let's say 700 words or 500 words how do you ask it to prioritize which points because in essay like you can add anything but how do you make your essay very impactful you tell it that out of these 20 points give me the top 10 rank them from 1 to 10 in terms of the relevance to the topic. Right?
What I'm doing is giving it constraints and I'm asking it to force itself to rank on a particular scale so that it goes from being vague to actually giving a logical reasoning. Now the best one out of all of this is the inversion principle.
Ask it to comment on where this will fail, right? Or challenge the AI.
Basically finding the failure in what the LLM itself gave. Let's say it gave you a solution, right? Ask it where do you think this will fail? You will have interesting answers. Try this next time you use uh like any LLM. Ask it to critique itself or ask it to give you loopholes in its own answer.
Yeah, I think this so now let's open your laptops and uh I think all of you have claude.
Yeah. So what you will do is you will chat with Claude and do the problem statements. You can take inspiration from your uh classmates or think of any problem statement. What we want to do is use all of these principles and come up with a one pager that will give you the solution of the problem you chose. Let's do this.
>> So some of you raised hands or nodded that you don't have claude. If you don't have claude, you can use Gemini.
>> Yeah. Yeah. any any LLM right I think the free version should be okay >> so we'll take 10 minutes uh 5 to 7 minutes for this one where the output of this particular section >> everyone please open your laptops you have to actually build this out use these principles start brainstorming with uh the LM on a problem statement you chose.
>> If you get stuck somewhere, you're not sure, just please raise your hand. We'll bring the mic and you can ask your questions. So, make this interactive.
You don't have to do this on your own.
>> The point is we want to help you through this thought process. But you have to actually do this, every one of you have to do this on your laptops.
Just let me know if you face any trouble or if you think you have some questions.
We'll see who will come up with a good prompt, who will come up with a good solution. End of the session.
If you don't have your laptop with you or you're facing any trouble, just uh like buddy up with your uh uh yeah with your friend.
So the outcome you want out of this section is let's say you have identified a problem.
I want you to show me that this is the prompts I gave. This is a chain of uh prompts and uh these are the three solutions or these are the three features it kind of uh suggested for this particular solution. So that's what you want. you want to get to the prioritization as well on let's say if I'm building a resume prep tool right so for that after this session I would have had three or four features that would be a part of that particular tool I want to build in this particular one two students that have started this prompting exercise, if you want to share what is the problem statement you have selected, please just raise your hand. We'll get the mic around the room.
If you're struggling to define the problem statement, you can just let us know. Just raise your hand, explain it to us. What is the problem that you have identified and we'll help you through it.
Students in the back, please don't talk amongst yourselves. This is step one that we are doing here. So after this, we are actually going to build something. All of you are actually going to build something.
And we will be not today but the process will be that we will actually evaluate your projects give you feedback and identify some top winners some top projects right so please work on this seriously Anyone who is ready to walk me through their prompt or their thought process.
So the idea here is not to find the best problem or not to find the best solution.
The idea is to kind of give you that structured thinking framework and how you can like get the best out of uh LLM.
So don't worry if it's a small problem or if it's just a silly problem that's not uh relevant here.
anyone is ready with their prompts. If you want to share, please just raise your hand.
If you need help, if you're stuck, please call us. We'll come around and we'll help you out.
the problem statement you've chosen the last solution >> sir actually um most of we use uh uh chargy for uh coding purpose right and uh in uh most of the ID ids and like visual studio and pyam we'll be using the most of the codes from uh using prompts and chargit and we'll just copy paste in most of the cases and we'll modify them in in accordance to our uses usage so why can't we merge the IDI and the uh ID and this u AI such that uh we'll give the prompt and directly the implementation will be done through the ID such that the time difference and Um can what can we like the time gap?
>> Yeah.
>> How did you start and >> what did you use to get to that solution? Like what is the solution?
>> Basically I use this uh like ascent person. I simply use this >> example of what your prompt was for each of those.
I just gave that the I want to give uh I want to analyze how this how this will be checking through the through the persona like different personalities. So that how this will be working through them like for example the senior particip uh partition and next the student developer and next the tech lead. So the these these type of personas I have used and next coming to constraints like um these um when we are using these prompts um we these need to be implemented in a spontaneous. So we need some uh quick uh quick developers and like manual thing.
Next um in coming to inverse principle like uh the editor need to be outside the AI and the developer moves through the code and this AI will be generating the code and need to implement it directly.
>> No, I'm just working on it.
>> Yeah, just go to the solution because it will be useful in the next step. Okay.
But he brought out a good point that sometimes there's not just one persona, right? He checked his problem statement for multiple personas like the student or the coordinator and all of that. So it is possible that there are multiple personas involved in a single problem statement. So you have to think for each of them separately as well. Anyone else?
Good afternoon sir. uh most people uh lack the awareness of that difference between uh academic problem solving and interview style uh problem like uh if it mostly interview in interviews they mostly ask like real time fa applicated problems right so uh so I asked the cloud AI how to solve it it's on the way it's >> sir >> what is the like I just uh I just asked it how to solve it.
>> Yes sir.
Really?
>> Anyone else?
I'll be uh I'll be building an agent uh uh based on uh engineering. so that everyone can hear.
>> I'll be building a blameless postmortm agent. Uh we do works uh projects best and we execute our projects using codes.
Uh we simply run it and uh if we get some errors we simply blame human error. We simply blame as a human error. What I'll be doing is to build a blameless postmortem agent. Uh it has coming to those uh characteristics assigning persona means I'll be building a site which analyzes the server problems. Uh eradicate the code errors um and it will be solving all the errors.
I'll be injecting a prompt uh which acts as a reliability engineer like Google.
uh Google just shows the result but it solves it at the background uh and analyzes a problem and uh makes it look easy and better. I'll be giving a required constraints by giving up the summary or uh specified prompt based on that. I'll be giving a perfect prompt to that so that errors doesn't exist. I'll be specific at every single thing when it comes to a project. I'll be specific at every task so that it doesn't it gives me a perfect output.
>> That's actually a very big problem uh in uh like website or SAS companies. So just try to fine-tune the solution.
We'll see how that prototype will look like.
>> Okay. Just share your problem statement on how you >> So basically the problem statement was AI resume selector. So my prompt is like uh you are a team of like front-end developers, back-end developers and also HR of top M andC companies who have strong knowledge about recruitments, ATS systems and professional resume building. I hope all of you know what is ATS system. So basically it is like a resume checker of like companies they identify through the use of keywords what you what you have written in that resumeumin and all. So it like the features would be AI resume suggestion system. So generate resume content according to the skills, experience, company role, industry. Then other feature is ATS resume checker. So implement ATS based resume checking system. Check for ATS scores, missing keywords, formatting issues, weak sections.
And there is other feature as well upload feature. So users should be able to upload their own réumé or the uh sample rum which can they refer to then edit their details manually then uh it should must have humanized rum writing like most of the students like who don't know about how to create réumé and all so they usually create from AI but they actually could not like distinguish which is better and when HR uh usually check these all things they can identify it is like completely AI written so they must have like a feature of humanized way of resume writing. Then uh the framework which had been like I mentioned in that ReactJS, Tailwind CSS, modern UIX and uh back end of NodeJS and MongoDB these all things. So uh in the conclusion part I wrote first give me complete project architecture explain the front end backend folder structure give me a detailed PPT style walkthrough of like how the website would be all the pages and output of that sample resumeums how it can be. So thank you.
That was my >> that's a very detailed approach and uh I'm going to pick up on the last part you said. You basically asked the LLM to create a PPT of how this tool would look like and how the flow would look like.
Next slide. So what you're essentially doing there is creating a let's say a workflow diagram for your solution, right? you you go and show it to someone that this is how this tool will work and this is how this uh particular uh automation will work. So that's where prototyping is very very important. So anyone heard about prototyping like I know professor uh Jagat covered it but prototyping is basically think of it like a dummy working version of your idea or a tool or a project right next so if you look at it so there's I've already thought about the solution for this right now that's the solution that's the workflow on the left side what do you get out of it right it looks very complex so for your use case what you're asking is create a PPT So that's exactly what I did. I created a PPT but no one understands what it looks like right or no one gets a sense of what exactly are we talking about. Now this is very important and one thing AI has done is changed the game in this particular uh area. Basically if I had to 5 years ago pitch an idea to let's say a VC or a CEO I had to go with like 60 pages very detailed document on what works what doesn't work. he has 5 minutes it doesn't get conveyed but today with AI I'm able to build a prototype in two three hours and give that look and feel of the product I'm talking about build it take it to them in 5 minutes I can demo it so that's what we're going to do now with lovable so how many uh here heard of lovable good so what we'll do now is use your problem statement from claude right you were trying to brainstorm with it now What I want you to do is build a very lightweight prototype. We don't want very detailed one, right? Build a very lightweight prototype for the solution you were envisioning, right? Ask Claude for a prompt that you can put into Lovable to create a prototype. So, let's do that. Whoever's still in prompting, that's okay. After the session, we'll have uh some more time. We can do one-on-one uh as well. But what I want you to do for the next 5 10 minutes is think about the solution, fine-tune the prompt and get that prompt from claude and put it into Lovable. Again, log into Lovable with your Gmail or any account you have. It's free and I think for this session it should be okay. So let's see if we get some interesting uh Lovable prototypes. So what you need in Lovable is just a prototype. It's like an instruction, right? Just take that instruction from Claude, put it into Lovable and then you you should have your uh prototype.
And even if you don't have a very detailed solution that's okay just to get a feel of how lovable works just uh like think of very simple website or anything you want to build for this section and just get a pro prompt out of claude.
If you have any questions, need any help, just let me And don't worry if you're not able to figure it out. We have uh another 1 hour after lunch where uh we can just like brainstorm about problems and uh try to build prototypes.
And if anyone's ready with a initial prototype or something just raise your hand.
>> Anyone else wants to share their problem statement?
You have to volunteer otherwise I will ask you to share any problem statement. See you're all working on your laptops right? So just don't worry about the perfect prompt right now but two or three students if you can share what is the problem you identified and how you're prompting Anyone from the back? Anyone wants to share?
Yes, we have uh one student raising a hand there.
Uh so I have given a prompt like we have a problem statement. This is very common. We have any number of platforms like whomever they want they can post anything. So but we don't know as a public point of view we don't know whether it is real or fake right for example as a student point of view uh while searching for internships or some course related to that we will not be uh sure that whether it is real or fake because uh there are many scammers they are there so we don't know. So I have given a prompt that uh be a top cyber analyst point of view and give me the solutions for uh how to solve this problem and it gave me a solutions like uh uh use this kinds of tools and uh such kind of stuff it gave then one more prompt I gave is be a hacker point of view be a fraud point of view so we will be knowing like for example I have heard like it's very common right in our age we used to watch cinema and that they used to say if you want to become a police thing from thief point of view.
So I have given a prompt that be a fraud point of view and give me this uh how will you manipulate public. So it gave me the solutions uh like uh I will try to make emergency I will make the create urgency like uh tomorrow like uh they will say to student actually I have experienced this in my uh nearly third um month of my first year. They have called me and they said uh like uh this is the last day for you. you have to apply for the internship. What I did, I called to my counselor and I uh made a conference call and he uh dealt with them like how can you say that you're calling from our college? So you can't force our student uh for your uh to apply for your course. So we can um like we can deal like this. Thank you.
>> That's a very good problem statement actually very interesting and uh very interesting usage of inversion principle as well. So what she did is flipped the persona and said now think like the hacker or think like a fraudster. So try to build a solution on that and try to build a tool. How would that look like rightly use lovable come up with top two three features you want in that tool and build a prototype. I would like to see that.
Anyone else?
Anyone else this side in the back?
Anyone student would like to share?
>> Anyone ready with a prototype or at least the first version of the prototype?
You can keep refining your prompts. So we will have a break. You'll have your lunch break and then we will again join here for 1 hour 1 hour 15 minutes.
Right? So that's the time for you to ask questions. So think about it, refine your prompts and start working on your prototypes.
But please ask questions. If you're stuck, ask questions.
The point is when you all walk out of this session today, you should be confident that yes, I can articulate a problem statement and frame it the way prane explained, right? I want X to do Y in that format. So here I want to solve X for Y that does Z. You should have to have a clear problem statement and then build a prototype.
You all should be comfortable to do that by the end of the session. That's our goal.
So my problem statement is that every year urban communities as well as the rural communities they face disasters right and government take actions after the results have been taken. So my solution to this is to build a prototype that will presimulate this uh disaster and help the few people's die in the result of that.
It the solution could be that uh it can be the type of area the number of houses the uh the number of people the the uh height of that place from the sea level.
These are the factors that we can use and also the organizations and governments can use the solutions to filter out their uh pre-built solutions and also they can help their parties grow, right? And so that's it.
That's a very interesting use case. In fact, it's like one of the very good uses of not just AI but predictive modeling, right? So, essentially what you're trying to do is predict the disasters and kind of reduce the fatalities and the damage that's going to in fact like we have like in one of my previous experiences, we actually built a dashboard for that where in US we wanted to predict hurricanes, right?
Because hurricanes is a big thing in uh US. So you have this data somewhere. The metrology department has that data. How can you use that data and track where that hurricane is going to go and see what stores are in that line and ask them to stock up more on those essentials. So yeah, but how would you go about building an initial solution and how would the dashboard or a model look like? I would like to see like once you build that uh prototype. Yeah.
Anyone else?
We'll just do one two more and uh yeah actually my main problem is that unemployment uh so before um in olden days they should there was one paper newspaper in that at the last corner every um like whatever the jobs are there they used to post on the paper and they used to call that like I want this job but now there There are so many social platforms like LinkedIn and so many and also some people are posting on Instagram stories like like I need I need an graphic designer I need an senior web developer like that. So now I have um it's very difficult to find on all platforms and see so I want to build an EI which actually sees all the jobs and gives the jobs which I'm actually searching to me this is my problems.
And did you start?
>> Yeah, actually the AI will go through all the all the platforms and uh it will like based on my skills. For example, if I am good at editing, it will search editing jobs and it will just sync on both of them and it will give solutions give job suggestions to me based on that.
>> Not it >> that's okay. Just find out more on that.
So this is a use case for what I would say something like a web scraping or a crawler kind of thing, right? What it does is the agent you can think of an agent which scrapes no LinkedIn and all of that and gets that job postings for you.
We'll do one more if anyone wants to share. Yeah.
>> Okay.
So basically what she's saying is instead of uploading a textbook or a PDF or a file and me asking hundreds of questions without me not clearly knowing what I want I want the model again the problem here is you're hitting limits right after 10 prompts it says like yeah this is something uh I cannot answer you have uh like hit your limit instead what you're saying is what if there is a world in which this tool has all the context it needs of my textbook and is intelligent enough to preempt what question I'm going to ask or what kind of answer I'm actually looking for and give it in a one prompt or one kind of answer so that my limits are not like exhausted right okay so >> that's A good point. So what she's saying is instead of random questions and answers, the model should know that this is what I'm looking for. So basically what you're doing there is giving it context. So this is an important topic in uh like the AI world context engineering it is called right?
So you are giving the model that this is what I'm looking for. You're giving it clues. You're giving it boundaries.
You're giving it restrictions that hey this is what I'm looking at. So keep the answer restricted to what I'm asking about. Right? And one topic which might be of interest to you is rag model, right? Where you're saying I'm going to give my textbook answer only from the textbook. Don't think about anything else. So that's a very good use case for a rag model. So read about that.
Okay. I'm just going to give you a quick demo of what I built using proto uh lovable.
So I had the same idea when I was coming to the campus that building an interview prep agent or like solving for the interview prep problem right how do you build your resume how do you prepare for the interview right again how do you prepare for aptitude test so this is a prototype again this is not a working model that's not the point of prototype I I wanted to show you guys what my idea was and prototype is the best way to do that. So what I built here is a interview prep. I'm just calling it prepper. So what I thought about in terms of features is for any problem think of the entire value chain. When I say value chain, think about all the steps that are involved in an interview, right? Or think about all the steps that are in involved in let's say choosing a career, right? Think about all the steps that are involved in preparing for an exam. That's when you can connect the dots. Right? For example, the way I thought about it is okay, I want to build an interview prep agent which will help me with interview preparation end to end. Number one, I wanted to review my resume. Number two, I know that lot of uh companies ask for an aptitude test. So, I want to solve for aptitude prep as well. And three, I want role specific technical questions as well because if I'm sitting for a SD role, my questions will be different for someone who sitting for a data analyst role, right? Third, fourth, there might be a group discussion also. So why don't we prep for a group discussion? And fifth, of course, the personal interview. Now personal interview is more on your profile and the kind of questions they ask based on your uh role as well. So just let's quickly look at what I thought about the prototype. So for ré builder, what I'm going to do is let's say I'm going to upload the JD, right?
Each company floats a JD. I want to fine-tune my resume for the specific JD.
So that's one feature I wanted to add.
You upload your resume, you upload the job description. It gives you what is missing from your uh resume. And that's where the ATS tracking and all of that also comes into play.
Then I also wanted to add an aptitude prep. So give me practice questions daily. Right? Again this is just a very simple question here. So just ignore it.
And uh role specific questions as well.
And in fact what you can also do is now that the technology is improving day by day maybe you can actually have mock group discussions right. So I thought why not include the voice uh mode as well because you can actually chat with uh like talk to chat GPT and get your answers as well. So I wanted to include that too. And of course the personal interview also what I envision for a personal interview is I upload my resume upload the JD it asks a very specific question related to that particular role and my JD. So that's how I wanted to simulate. See if I would have shown you that flowchart and explained you it would have been difficult. But now you can see within two minutes we have covered what my idea of this particular product is going to look like. So that's what becomes easier with prototyping.
Any questions? Any thoughts?
Cool. So we'll break for lunch Kushbu.
Yeah. Okay. So what we'll do is after the lunch we'll come back and we'll start building, right? We'll in fact I wanted to cover how you can build it with cursor. I think one of the problem statements was actually uh having an ID and building within the ID itself. So cursor is more or less doing that. So we'll we'll see.
All right. So students now we can break for lunch. You all have to come back to the same seminar hall by 1:50.
All right. It's 12:45 now. You have to come back by 1:50. We will have another 1 hour where we'll do the rest of the part. So right now >> we have 1 150 50 right not 1:15 10 minutes before 2:00 what we are going to do is we have only done the prompting part so far some of you have built prototypes we are going to finish our prototypes and then move to cursor right we want to make sure by the time we finish today you understand the entire process those of you who have laptops that are running out of charge please make sure you also charge charge your laptops when you come back.
>> So we'll be taking attent from 1:50 onwards and session will start short by 1:55.
Those who have classes in the afternoon session, the faculty will come here and they'll mark attendance in ERP.
I'm repeating those who have classes in the afternoon also come to this room. your faculty come here and they'll take the attendance.
Thank you.
>> Thanks everyone.
But uh so here's what we'll do after the session. There were a couple of folks like students who came to me and showed their prototypes and in fact like there were really interesting ideas in the room. So what we're going to do is call up some of you and uh we'll give you a chance to like present your prototype and walk walk through your uh solution as well. So we'll do that while they're presenting have a look at what they're doing but also keep working on your prototype keep working on your prompt as well and if you have any questions right and if you need any help some of them wanted to fine-tune the problem call me uh Kushbu we'll uh come sit with you and we can fine-tune that but uh I saw some of them build uh some interesting prototypes can one of you uh who came to me earlier before lunch come and present your solution ution uh here someone here had an interesting prototype and uh >> you also had one right yeah bring your laptop so what what we'll do is couple of folks like three four people we'll take two three minutes each just connect your laptop present your solution here and just walk your uh uh classmates with through the process you have followed >> that's a good idea Okay, anyone else? I think you also built something, right?
Okay, while your lovable prompt runs again.
Again, everyone else, start prompting, start building those prototypes. And if you have any questions on fine-tuning the problem statements, call us or else you can ask the LLM as well. So, I think it takes time to like reload.
>> No, no, I did not connect.
There were three four students with very interesting ideas and uh I saw the prototypes as well. So please identify yourselves, come up on stage, present your prototype and your uh problem statements.
But rest of you start prompting start uh working on the prototypes.
You have the prototype ready.
Come on.
As I said earlier like uh my problem statement is like uh we have any number of platforms and we don't know whether it is real informations or fake information. So I have uh created a demo um web page. So which shows that we can uh paste some link any of the link here and it will show the risk rate. For example, we can take any link uh here we can paste the link and will it analyzes it whether it is real or fake and it will uh analyze the risk amount like how much percent it will it is of risk. So here is the demo of that and this is also it's like that you can drag the photo or place the same link and uh analyze the target you can analyze it.
>> That's cool. Uh can you show your output once and just walk us through what that output is?
See guys, if it was just a piece of paper or a flowchart, it would have taken her 10 15 minutes to explain. But in 2 minutes, I can see what she's trying to do here. So that's the power of prototyping. And with everything that's happening in the AI world, it's going to get more and more easier for all of you to build apps as well, right?
Again, not just software engineering uh folks, but even a chemical engineer or a fashion designer can build something on their own.
Can you explain what that output means?
>> Uh so here it uh explains that risk percentage out of 100 how many percent it has of risk. So it is of nearly 12%.
So which means it is less risk. So it is safe. So of course you can believe the content of that and uh like you can follow that like you can believe that.
>> And did you ask for it to give a specific score or uh did it uh give it itself like how did you decide that okay I want a risk score or I want a verdict?
Did you give it in your prompt or it was a suggestion from the LLM itself?
>> Suggestions from LLMs.
>> But yeah round of applause to her. She's like very interesting solution here.
Now the thing is this is just a prototype right? What you can do is get some basics on cursor. Uh Kushbu is going to share a link with all of you on a course. We we have curated a very specific course for you which talks about prototyping which talks talks about prompting and also building with cursor. So once you finish that course try expanding on this idea and actually building an app that works because right now what she built is just a prototype.
It's just showing you how it looks and feels like. So you should do that.
>> Thank you. Thanks.
Anyone else?
Who else is ready? I know there are like couple of folks who have done some really good prototypes. So, please identify yourselves, come and present your solutions to your uh classmates and others. Keep continuing prototyping and prompting and if you need any help, if you want any sample problem statement to try out, we can do that as well.
Someone had an idea of I think anti-LinkedIn. Everyone shares success stories on LinkedIn but someone was having an idea of what if we can share failure stories and learn from them. I forgot who it was.
What did that?
That's a good idea. Don't worry. Uh I can help you with presenting that solution. But I think it's a interesting solution. Everyone should see.
I know you are in the room but yeah I forgot.
Anyone else ready? Anyone has questions, thoughts, ideas?
It's loading.
Okay, we have one more.
Hello.
So this was the resume website which I made uh using the prompt. It's not like complete. It's like a dummy version which uh like it doesn't have currently any data. So I'll go through the website once.
So this is like interface here like signing options and these all things are there as you can see over here.
Then here's the template section. Uh like right now no templates are available. So it gave like a decent uh like all that earlier it cames as default uh template like you can see summary, experiences, skills, educations etc and etc. There are other types of uh also data like for different roles also it is given. Then there comes like uh section of ATS checker. So this is the most important like uh feature of this uh website would be which will check how much our resume would be rated for that ATS feature. So we can see that uh for the dummy uh template of that ré the score would be given like how much it is given. The more the score the better the rum would be. So uh you can see over there the target role is assigned like for SD in which company you want to apply the job descriptions and all as well as uh how should my resume would be. So based upon that uh it will run an ATS analysis and it will also like uh give us like uh suggestions like what should we keep what should we avoid and majorly like uh last uh last week only I was building my own resume. So that time I me I mentioned some keywords. So basically when you are giving your resume to an HR so they don't check like whole resume they don't have like time to give uh to check all this resume. So they look for keywords that attractive keywords which will make them like grasp their attention in a just a span of time. So like you can see the keywords like distributed systems, Kafka, AWS, microservices and etc and etc. So these keywords if you mention in your like réumé it will be like much more preferable.
So here's the dashboard section like you can see how many resumeums it's uh uploaded over there best ATS score upon like how many rum you analyzed the last edited version how much how what it was and here we can also find the list of rums for like different roles like uh for suppose uh I applied for an internship uh regarding AI automation engineer but along with that I also applied within founders's office role so my like my resume would have to differ little bit according to the role. So we can keep like arrange our resume according to that also. So here's these all. So you can see that was all. Thank you so much.
>> That's really great. Very well done, very well thought out and the level of detailing is good. I think it will be amazing if you can build something using cursor and deploy it again of course on your laptop. I think it can become a very important uh tool for you to actually use and like prepare your resumes as well.
>> Yeah, sounds good. I'll definitely try it.
Who else? Yeah.
Uh good afternoon everyone. Firstly I want to say that uh this don't have a fancy I mean the landing page and etc. But I more focused on the logic and the LLM that I used. So I have used the cloud LLM right here. You can't see it because it's uh not like we can't show that etc. So here is the model. I have used the set 4.5 for that. And here is the whole uh project that I did. So first is resume versus job validator.
Yes.
So here is the resume versus the job description validator. Let me show my resume for you all. So here is my resume and let me just upload it over here.
So here it is.
And for the job description, let me just upload a text file that I have.
So here are the samples that I have created. You can see the three réumés and three things. Uh so I have just prepared them for the sampling. So I I'll just uh click this over here and then I'll just click open and let me just click on analyze. So here it start analyzing.
So it takes time because this is not a flashy just a dummy uh prototype. It actually analyzes the core uh idea core uh functionalities of your resume and the job description. So it takes time because the LLM should load and things like that.
>> So you're actually built a working prototype. So what's the model you're using here?
>> Uh clawed sonnet 4.5. How are you calling it?
>> With the API.
>> You have the API key.
>> Yes, sir. Okay, that's good.
>> I think it might take time to uh give the output, but why don't you just walk through different features you have about?
>> So, in the meantime, let me just go through the dashboards.
So, here there is a resume versus job description validator and job job description aware rumé customizer and here is an aptitude trainer and your technical MCQ. Let me just open the technical MCQs and the rest two are like I was not at uh developed them. So, so let me just click on SQL for example and then click on generate. I have kept the difficulty level as medium and the questions are 10. Let's wait for a sec.
So what he has also done is gone a step beyond building a static prototype. He also added an API call for LLM. So this is actually a working prototype, right?
So there's a static prototype which will just give you idea of UI and this is a step above that which is a working prototype. It still gives like valuable outputs.
And after this, the next step is building a fullscale app and actually deploying it on cloud and sharing that link with someone else so they can also use it.
We'll wait for the output to be generated. But uh thank you so much. A round of applause for him as well.
Really, really well done.
Anyone else?
What are the rest of you doing? Are you prompting or just looking at what your teammates have uh classmates have built?
Any questions?
Can you answer them?
I see a RCB fan here.
Hello everyone. Uh I'm Shavi. So let me tell you what my problem statement is.
So firstly we all are students we have AI tools many things to get help but if we uh if you think about farmers there are no one to help them mainly everyone gives false assumptions to the farmers cuz the main reason is they are not educated right so that's why I I wanted to think only about the farmers so I wanted to build an app which helps farmers so my app uh name is agree AI.
So this is my prototype. So there are uh mainly if you come to the problems like the uh farmers don't know which crop to uh like how I can say if uh based on the soil which crop they should um sir >> yeah basically based on the soil what crop uh to kind of sew.
>> Oh yes sir. And also they don't know how to use the resources like fertilizers.
You all know about the middleman. We all know right? So what they did means they don't the farmers don't know which fertilizer is the best. So the middleman will do will give for more price. So the farmers can't afford it. So that's why my solution will will give the suggestions of some fertilizers which are which available fertilizers will help them and for low cost they means and also they don't know how to use the fertilizer. If they give more fertilizer to the crop, the crop might damage. And also if they give low fertilizer, the crop may not work. So that's the main point. And also when the rain will fall, when the crop will damage, they don't even know. They can't directly check the weather updates in their phone, right?
So in this app, it's also there like weather updates. You have show I have already shown in that uh first dashboard, it shows that heavy rain. So what they will do at that side heavy rain alert. So what they will do in next 48 hours postpone fertilizer spraying.
So means the suggestions also they will give means what they have to do when uh rainfalls. So some suggestions it will give and also pest attacks. When the pest attacks they immediately don't know the solution they run away to some other shops. So so this is my pest. So what I have to do now? They'll think like that.
So that shop shop shop members or any middlemen they'll take it as advantage and uh they'll uh sell something something else. So that's the main disadvantages only for the farmers. So in this it also gives the best suggestions also. In here uh they they just have to upload the photo of their uh crop. Let me show you that here. If we take the photo of the crop, if any pest detects, it will give the suggestions also. Uh you all get the doubt that how it will how every farmer reads that in English. Uh many farmers don't know English, right? So they can also see in Telugu also and also they can hear in Telugu also. If we uh here an interaction button is there tap to speak means former directly interacts in Telugu with AI assistant. Okay. So that the former will easily interact with AI in Telugu if they come from north they'll uh easily speak in north language everything right and then soil scanner it will scan the soil and also fertilizer suggestions it will give as I told earlier and also market prices so this is the main thing actually the farmers don't know the market prices so what the persons in the middle do They'll uh they'll what they'll give just low money and they'll sell for more price in the market. Who is the uh means what is the disadvantage or for whom it is? It's a for farmers everything for farmers right. So for that they can know what the dealers and what the middleman can do here. Here itself they can see the market prices everything and also here if you see the interesting one report dealer. If the dealer says uh for uh low cost what will happen here it shows high cost. So he can the farmer can directly report the dealer. Okay you can understand right. Next government schemes here uh I will update it later. Here all the government schemes about the farmers they will know to get uh in uh money or else any water any fertilizer help from them they can know easily here itself and also profile everything what they did how many crops grown every detail is there in the profile. So this is my total agree AI prototype and there are many many futures upcoming. So I'm just working on it. Thank you.
>> That is amazing like the detail she thought through. So this is what problem solving is all about. She actually like went into the details of the entire value chain and what's your name by the way?
>> Shani.
>> Shavani. One more round of applause.
This is very inspirational right. This is she did not just think about interviews, jobs and all but building for something which is very valuable. So what are all the challenges he will face in this?
I'm here.
What are all the challenges you will face in this?
While building the app sir.
>> Yeah sir. Uh actually farmers in main villages they have no internet. So I just want to make the app in offline mode. That will be the main challenge for me. I guess sir.
>> And another thing is uh you are interacting to the app using voice or speech.
>> Yes sir.
>> And uh there will be a problem with accent and dialect.
>> Yes sir. How we will take care because whatever the farmer speaks >> it may be in a collect language even they may not speak even telu in the proper way.
>> Yes sir.
>> Uh how it recognizes is a challenge. It is a very >> of course it's a challenge sir. I'll I'll work on it for sure sir.
>> Thanks Ravi. And uh who knows maybe in five years you will build this app and like you will become a startup owner but uh really cool stuff.
Who else? I think she has raised the bar. But again, that shouldn't deter you from sharing what you have.
You want to show the output? Yeah. Plug in.
Sorry for the late response before. So here is how it looks after the analysis.
So it will show us the match percentage about your resume and the job description about the ATS friendliness it is. So it shows the ratio etc. And also it gives you some remarks etc. It will analyze the key words that you kept in your resume across the job description which will be more useful and it will say missing all these things and the improvements that you should do and coming to the next thing it is here the technical MCQs for your job prep etc. So it will give a set of questions like this and uh this will is a add-on feature. If you give a job description and a present resumeumé, it will update based on our job res uh description etc. So here I gave my old ré and a job description. It just updated my resume and we can download it in a docs or in a txt format.
So here is how it looks after the updation.
So it'll update all my skills, rearrange my expertise, uh rearrange the keywords, etc., etc. And this all works in a cloud cloud. So I'll be pushing this without the API and then if anyone want to use they they should subscribe it for this and etc. >> Yeah, >> that's all.
>> Thanks for bringing that last point up.
So you can build tools and someone else also can use by making it uh modularized one and other thing is like you give them an option to input their own API key, right? So maybe this is he can host all of that code somewhere on versel or lovable and then you can download that and then host it in your local laptop and give in your API key as well. So yeah maybe who knows in next couple of months he can build that full scale and you can already start using it for uh >> so for building this I have you used a software called open work so it's a free open-source thing so we can use that and build the things up. Thanks.
>> It has given some suggestions, right?
Yes sir.
>> How many suggestions are acceptable? How much percentage roughly? Can you accept all?
>> Sir, it just gives us suggestions and we need to undermine the things and we need to update or else there is a feature it auto updates and gives us the downloadable file. So we can download it the resumeum.
>> I'm open to take any questions.
Yep.
Hello.
>> Hi. Huh. So coming to your website, do you think like everyone is like economically feasible to like purchase an API key to input directly to get that obtain that feature because like a minimum API key cost like $5 and along with that GST it might be like five 5.5 cuz I have already like purchased three to four times and for a normal person nobody would like to like uh to access any website they will go for like paying for API keys and all. So what is your views regarding this?
>> Okay, uh that's a very valid point. But coming to the thing right now we are very much focused on the prototype where I going a step extra and done this. So we can also make this open source with the Gemini API which is free for all or else some other things I going to look into it and yeah it's a great insight that I get. So uh one piece of information for you like uh just a small advice like rather than using like uh API why not train your own model and you can utilize in that rather than like uh using API keys for much more production at as it will boost boost up your like cost funings and all.
>> Yes. So here we are doing a semantic search for the resumeumé and the job description. So for semantic search we can't write our own Python code or any code which really makes a very good semantic search. For that we need some LLMs like Gemini or Claude or anything.
So clude that I found >> but my only question is like everyone is not like economically feasible that they can purchase a $5 uh API key and use >> that's the first question you raised and I gave an answer that it's just an prototype right now and I will enrich my things and I'll make it feasible if time permits. Yeah, >> exactly. You should do that.
>> No, but uh the lesson here is like there are multiple approaches, right? Some of them are open source. Some of them are leveraging existing uh LLMs and their APIs itself. So these are the things machine can't do. These are the decisions you have to take. So yeah.
>> So any other questions? That's it.
>> Yes sir.
>> Yeah. It's not easy to develop an own model or train a model. So >> thank you.
Anyone else?
How many of you are actually working on prompting and not just looking at something else on laptop?
Guys, this is for you, right? If you have any questions in the session, ask get it cleared within the session itself because we are going to float a course and we expect you to like complete three projects in that particular course. you want to touch on those?
>> So to all the students, we have all of your details, we have all of your email ids, you will all get invited to a course on Corsera and you are expected to complete that course. As a part of that course, there are six modules in that course, right?
So three of the modules have a reading item which includes a space for you to share the link to your project that you'll build. So what we are doing today is just a precursor to that. We are teaching you how to do this so that you can actually do that course and submit those three projects in that course. Our team that's Pran and his colleagues in Cosera will be doing a detailed evaluation of your projects and we will also be identifying the top projects and felicitating those students.
So please once again we still have time we have about half an hour. Come up here present it. Don't worry about whether it's perfect, whether it's right, whether it's wrong. This is your testing ground right now.
The fact that you're here and you're building something that is wonderful in itself.
This is the time and space for you to make mistakes. So please come here, present what you're building with confidence, right? And we'll tell you how to go to the next step.
Come on. We need two more students. At least two more students who are at the prototype stage. Please come up here.
Yes.
Very good.
The mic is still working though.
Hello.
>> Good afternoon. one and all I'm Sai one of your classmate uh my problem statement is about the civic issues we are facing day-to-day scenario it is an smart hackathon problem uh given by one of the north state government uh the problem statement they briefly discussed about is the people who are facing the issues the government need to give some opportunity to the citizens so that they can upload and give the perfect location to the government uh in order to resolve the civic issues. In day-to-day scenario, we can see that the person who faces the civic issues uh he needs to go to some administrative government block and then he need to report the issue.
The citizens need to face the citizens who are facing the civic issues need to go to the administrative mun blog and they need to mention the report like here there is a uh path holes issue at the perfect location.
Sometimes it makes late that the government needs to take the action and we don't know that if the action is in progress or it's completed or it's not started. So in order to increase the transparency among the government and the citizens, the government asked us to build a website that we need to give the input location and the camera access to the citizens so that they can upload their issue and we came across that the some of the citizens has uploaded the same issue. So it may becomes complex for the web. So in order to reduce it we came across with an idea called upport.
So that the citizens who are facing the same issue they can upload. So that the government with highest the post with highest ups the government will find the issue and resolve as fast as pos.
But uh this is a very good uh problem to solve, right? Very good uh governance problem. Round of applause to him.
We'll see his prototype once uh the presentation is back. But anyone else?
Okay. Apparently the internet is working so you can still work.
Backbenches are not getting the Wi-Fi it seems.
Um all the students who came to the stage and presented can you please stand up? Pran is just going around. Um we'll hand over the kids.
We had five students come up here so far and present their prototypes. Can you please stand up or raise your hand?
All right, everyone please listen. One more student is here.
>> Good afternoon everyone. This is Vak. So uh we are generally we have don't knowledge about the law and sections of the in our constitution and we are difficult to understand the what is the laws and what is our rights. So I have came with an idea which is an AI related. Uh first suppose we have a book case like and which is uh doesn't we have done. So what is the purpose and what is the case must be and uh what is the correct process in that? So it develops like you know rights articles and cases.
What is the F So it is the chart box right now. So I was giving in a small case.
So it was giving like an immediately contact your uh I was given a case like a person was stolen my wallet. So it was giving a high priority that immediately contact your banks to block all the debit cards and what is the sections in that and it was giving like in past cases it was solved within.
So uh it gives you in what is the sections uh we have to file in the cases and it gives you in past cases like there is any past cases we must uh file in the history also. So what is the high priority we have to done uh and I have like rights and articles cases fire agency and help lines.
So uh it gives like an in F how to file an FR what is an F it was a very important to know what's our rise in the society and how the people was democratic to know how we live in in India thank A round of applause for him. Uh thank you. This I think this is an interesting uh problem. Lot of times we are caught on the road right without helmets triple riding.
Probably this will be useful there.
Okay, we have one more.
Okay, what do we have here? So actually I already said that I want to create um a agent such that um we want to make uh the co the coding which we copy from the AI um chargebt or some AI agents we can directly implement them uh using these agents like this agents for example uh we use uh for fullstack development we'll be using the ids for coding right and then the implementation will be done and by using like This one by giving the direct prompt it'll be displaying the output where the here where here we'll be doing the code where the code will be automatically generated and we can modify in it for example uh I am entering like build a login page authentication.
I think it's going to take some time.
Yeah, we'll present it once uh the screen's back. But uh good job. Yeah.
Guys, a round of applause for what's your name?
>> Koshik.
>> Koshik.
Unfortunately, like we're not able to see the prototypes because of the presentation, but uh I think we have one more. We'll uh just do this one. Explain what the problem statement is. And uh again folks uh I would love to see all of the prototypes you've created. So just share the screenshots. I'm asking someone to go around and collect some screenshots. It'll be great to see what all you have built. Again, the course has a option to actually submit all of these projects. So, we'll definitely validate and like score all of these.
But uh yeah, good afternoon everyone. I'm Ban Tarunal Anjin Katka. Today my problem statement is that we'll go through many of the certifications that we are going to do in the next four years. So we don't even know what is the guided flow from go to a course to course. So you'll be having a chain of links that will get you certificates and that will make you get hurt. Even Corsera is a platform that will provide you many of the certificates. So I wanted to create a website that could solve this problem.
So that if you want to become a UIOX designer or a software engineer, you'll get a guided flow of certifications that you have to do to get a certificate job in a top tier one company.
The packages are not involved in this since most of the packages are demanding only based on the skills you are going to have and these skills will come from the practical implementation of the certification that you will do not from the passive learning or the just for the sake of completion. So you have to do each and every certification with the dedication that will get you the skills and get you hide in the top tier one companies. So the website that I'm going to create will solve the problem of passive learning and will get you the sets that will really get you hide.
Okay. So while he's connecting everyone please pay attention. Now what is going to happen as a next step is by tonight you can all expect an email from Corsera inviting you to a course on Corsera. All right. You'll all get you'll all get an email saying you have been assigned this course. The name of the course will be masterclass on AI first product building. Exactly the title of our session today, right? It's a two to two and a half hour course. It's a custom course. It is not a standard course from the catalog. It's a customized course that we have built specially for you. It covers exactly what we have covered today. So it will cover first prompting using any of the LLM, cloud, Gemini, whatever your preference is. Then you'll move into prototyping using lovable and then actually building your projects.
Please listen to me for 2 minutes carefully. You will be evaluated on this. So please pay attention.
In this course that you'll do on Corsera, there are six modules. Module number three, module number four, modu modules uh sorry not three, four, five and six. The last three modules have some videos and they have some reading materials. One of the reading material has a link where you have to upload your final projects. So these are prototypes.
We have done work up to the prototyping stage. You have to actually build a working model and you have to submit the links to your projects. Our team at Cosera is going to evaluate your projects thoroughly. It will not be just a one-way process where you learn something and you submit something and that's the end of it. We will actually be reviewing it and giving you detailed report cards. We will evaluate your projects, give you feedback where it excelled, where there is area for improvement. Right? So we'll be giving you a very detailed feedback and we'll also be identifying some top projects and felicitating those students.
All right? So this is the next step because the whole point is we want all of you to feel confident that you can build your own products.
That was the whole point of this workshop, right? that when you finish this and you do the course and you submit these projects as a first year student already you feel confident about building your own products following these steps.
Everyone with me on this? Can I get a thumbs up from everyone? Everyone understood the next steps right? We will be monitoring closely how many of you how many of you are submitting. So I will have a tracker. I will be able to see how many of you have submitted your projects and you will be evaluated on that and we will be rewarding the top projects. So please work on it sincerely. This is for your own benefit.
Once you master techniques like this and when when once you master uh you know processes like these it'll become very easy for you to start building your own products solving some of these problems on your own and then you can even mentor the first year students who are going to come in later this year.
Right. So, please take advantage of this opportunity. Look out for an invite from Corsera to this course called MasterClass on AI first product building. Enroll in the course. Complete the course and please make sure you submit your pro uh the links to your projects in module four, five and six.
There are three projects in that course.
All right.
And uh guys basically what's happening in the industry is along with resume now companies are asking for portfolio right for example if you're a software engineer of course you will have a hacker rank or a lead code profile they're expecting something similar for everyone as well. So if you are able to develop these skills and before you go to interview build a portfolio of sort right like these are all some personal productivity tools or any other tools or automations you've built create a portfolio of that and when you go to interviews share that link with them along with the resume it makes a difference. So you can use all of these skills to build a portfolio. So it's like a resume. Rumé is what you have learned and this portfolio is what you have actually shown that you learned right. So yeah.
So coming to the actual prototype, you'll have options regarding how you are going to get started as well as you will give the login details. So since this is an prototype, not all the buttons are functional and responsive.
So I'll just go through the landing page so that you might get a clarity regarding how the model works. So actually many of you will go through the Instagram and in your Instagram you will see videos at uh 7 days, seventh educations. These will get you hired and all of these things might get true if you really done this. So but no one knows what are the exact citations that you have to do to reach your goal.
This website is something that will connect you to your goals. If you want to become a UIUIX designer as well as the time you are going to all lot for a week or a day will make you choose what are the best certifications that will get you hired. Many of you will have your favorite embassies and companies that you want to work in and this is a way of path which will get you hired in your favorite companies.
So these are the basic certifications that are going to work for everyone.
Whereas these are verified certifications that will work for the professionals and at last you'll go through a test that will give you the quiz results which will make you accurate and precise regarding the certifications.
Thank you.
>> Thank you. And uh that's the end of our uh session if you guys have any questions and yeah you want to present we'll do the last one but again uh you guys can connect to me connect with me on LinkedIn if you have any questions you want to share something please do share okay few folks uh couldn't get the chance to present we'll let them present and uh then we'll So good afternoon. The main uh problem I have been solving here is unemployment.
So what it is is uh there are so many people who are lacking jobs. It's not that they're lacking jobs, it's just they're not able to find the correct jobs and for example I'm applying a job for software engineer I have so many platforms like linkin and fever and all so many uh for having so many apps I might get confused and there will be some like as she has said there might be some fraud jobs which are getting popular nowadays. So to to see actually jobs which are useful and which actually sets my interest. I have built an uh AI agent.
First actually it will chart with me. I will uh for for example I will show what I am interested in. Uh for example my interests are um coding and building projects and I want to be a senior developer for example. Then I will chat with the agent and I will tell that I want to be a senior developer for example. Then it will search all the market and I will also explain that what type of job I want. For example, I want a remote job for just 9 to5 job like that which is work from home. Then I will explain whatever my needs are and it will search all over the market and give my give my uh give the jobs which I'm actually interested in rather than showing all the crap on the internet. So uh let us go through the prototype.
>> Good afternoon everyone. My name is Rai.
So how this prototype works in uh daily life is here uh get started free here I'm chatting with this one. So here I wanted to give what are the uh interest skills and I wanted to which type of job I want I wanted to give here also and what are my interesting and what are my uh communication skills and here I wanted to give and uh this one is started with me if I want if I am a user like this studies mean so it gives some uh and then here is the dashboard in this dashboard my profile will be there which I before I charted in this. So here job job matches this one is based on what I uh I gave interesting skills in before that chart. So this is the job matches based on my interesting skills and and uh it will be applied here and like junior python developer and data analytics intern and front end developer back end engineering like that it will show and this is my resume based on my interesting skills. It will created a resume. So here is the uh my resume like uh it will download a PDF based on my skills uh and my interesting uh and my interesting based on this it will created a resume. So this is the prototype. So uh furthermore we'll work on this pro prototype. Thank you.
>> Thank you so much. Uh round of applause.
I know like a lot of you want to present uh more prototypes and I'm really excited and uh I wish there was more time. we actually have to meet some of the professors for another meeting. So what I would suggest is uh you can all connect to me on uh LinkedIn and you can share your prototypes and uh share if you have any questions and uh I would uh like be open to help you guys and direct you. But uh I had fun today. Hope you guys also had fun. You learned something valuable and uh please uh do reach out to us if you have any questions regarding the course. And uh yeah, thank you so much.
Anyone who presented and didn't get the uh the swag, just let me know.
So everyone please don't leave yet.
Professor Harikiran will come and give the closing remarks. So please just be with us for another five minutes.
You can connect with us on LinkedIn as well. So um pran do you want to for my profile you can search on LinkedIn as Kushbu Kumar Corsera you can search with these keywords you'll find my profile and prane yeah so you can search by our names you can connect with us on LinkedIn if you have any follow-up questions you can do that if you also want to do so if some of you are interested you want to have a follow-up session you can also relay this to professor Gopi um or to your administrators to professor Hari we can do another virtual session probably sometime next week like an office hour those of you who are interested who have more follow-up questions you can join the call on zoom and we can have a 1 hour session right so we can do that but like I said please look out for the invite to this course on Corsera you have to complete the course and submit your actual projects and we will review them we will evaluate them and give you a detailed feedback on your projects.
All right students, we have one more presentation. Prrenit, there's one more very interesting one that's different from the ones we have seen so far. So, please let's hear out uh Nikil.
>> So, I have also made another AI So I have made an another AI prototype which is called Karuna. So the if we if we found an any animal in injured or any something. So we can upload an image and also our current location like this. So or else we can also type the location that we are present in and also we can optional select the loca language and we have optionally describe the situation like it's bleeding or something of that sort. So here what it the report gives.
So we can see the status and the severity of injury and also we can see a prescription and an administrative guide. It will say that tablets and all something that we can give it to it and also the plus point is that we cannot actually go to an animal and give it a tablet or because we are not a veterinary by definition. So it will give some recommendations like you can keep the tablet into an egg. We can smash it and you can give it to it or something of that sort. So we can also create an images for which the uh we can frustrate like if we say I have an oil bottle how to apply it. So if it will give an image so how to apply it or etc. So we have something so see so we have nearby support we can find the veterinaries and also the other local helpers re near us and we can call talk realtime talk with something called sitha. I have kept the name very indigenous because of the app name itself is a karuna and something like that. So sitha analyzes the images and then it actually make us to talk like a veterinary person etc. So we can also like if you say I have an oil bottle and a cotton it will say how to apply it or we can ask some questions etc. So yeah that's all it is. I'm open to take any questions or insights.
So for the further thing I am also incorporating something called NGO network. So where the NOS's or others can interact there and also they can showcase the NOS's into something like PETA and something other more uh things so that the NOS's get the reach. So it's a social media platform for NGOs. I can say it as that. So that's all from my side. Thank you.
My very first thought as in I got into this room is I missed being with you all right from morning. I could see some interesting presentations happening. I keep getting the videos from the earlier presentations also. So it's indeed uh the feedback from Corsera team is your students they are highly interactive.
They have asked ample of questions.
They did not let us continue forward.
They asked some really brilliant questions, some really sensible questions. Right? That's that's a wonderful feedback to hear from a team like Corsera and that's a wonderful feedback to hear about firstear students right if you are a final year student then that's a different story but if you're a first year student and if you're asking these kind of questions I think that's a very good uh feedback that I have heard today and that that made my day and what I instantly felt was I should have been here right from morning. I should have sit with you. I should have seen all the presentations myself. But I've got all the videos.
I'll see them. Uh I'll I'll I'll ensure that I watch every video that I get. Now the next steps as Kushbu from Corsera is mentioning is that you get an invite to a series of courses or one course in Corsera. You'll have to complete that and you are supposed to submit. There are three projects in that. So you're supposed to submit those three projects.
They will give a very detailed report on every project and they will felicitate the top projects. Right? So they'll come back to the campus and we'll have one more session like this where we will felicitate all the uh students who have done a good job with with respect to these projects.
Right? So keeping aside that felicitation part, it is important we continue what we have learned today and we take up that course and we build those projects and as you build these projects discuss with your peers see how you can fine-tune them. You already had today in the class you had few students who could build some really good know agents build some really good you know products. So discuss with those students, fine-tune your product and ensure that it is part of your portfolio.
Ensure that you showcase that on your LinkedIn. Ensure that you showcase all these products on your LinkedIn and elaborate a little story about your entire experience, right? How did your day start? What did you learn? Who have taught you what? How did you build these products? and what is your experience throughout?
So once you write this kind of a story, once you build this kind of a story on your LinkedIn profile, students, I need you to focus. No talking. O once you build this kind of a portfolio and you keep adding your projects right from first year, I'm not just talking about these projects. I'm also talking about the projects that you do as part of your academics. Once you keep adding all those projects into your portfolio, by the time you come to third year, by the time you finish your second year, you're already industry ready. And if someone from industry sees your portfolio, placement is something that I should not be worried about.
And that is the aim with which I work. I don't work targeting placements. We work targeting building your portfolios in such a way that placements happen naturally.
Placement is just one outcome which should happen very naturally. Industry should identify you and they should call you and they should say that we want to give you this job. That's the kind of scenario we are trying to work towards.
Right? Not a scenario where we invite companies and say that you know these are all my students can you please test them? Can you please take them? No, those days are gone. They're slowly diminishing.
The future trend of placements is no more companies visiting campus, conducting a test for you, conducting an interview for you and selecting you.
They will look at your journey right from first year and they'll go through that through your footprint on social media, right? How is how are you building your LinkedIn profile? At what age, at what year, in which semester you worked on which kind of project? What was your learning? They'll see that entire history and they will themselves invite you or they'll call you for a hackathon.
They'll call you for an ideathon kind of a session and they'll ask you to work as a team with their own people or among yourself and then they'll choose you.
That's the future of placements, right?
And all that would naturally happen if you start showcasing your work.
I keep telling students always that it is extremely important we showcase our work. If we cannot showcase our work it is as good as we didn't do it.
And when you work, make sure there is lot of clarity, there is a clear goal with which you work and there is lot of honesty with which you work and there's a lot of hard work that you put in. With all these factors ensure there is showcasing also.
All these four parameters put together they will ensure that you are shining for placements.
Okay.
So let these sessions be inspirational for you all to build products to plan for your future to think way beyond what any other student of your age would think right think innovatively even with your course related projects don't get just stuck with the projects that faculty want you to do think out of the box think innovatively and come back to us saying that this is what I want to build Right? Faculty would definitely have their own suggestions because they have to deal with a class of people.
But you don't have to stick to that specific groove. You can always go back to them and say that I want to do something different.
So come up with such innovative projects and enjoy your stay at KL for the four years. I would only say three years.
Fourth year is in the industry, right?
So all the three years that you're staying here, make sure every day is enjoyable with a lot of innovative projects, right? And I'm sure you guys are going ahead in that direction only.
If I can see you all sitting in summer studying, that speaks volumes. While all your friends are on vacation, if you're here studying, that speaks volumes about the interest that you have, right?
And as I promised from next year your schedule will be in sync with the other students you would get a summer vacation. Okay. All right. Any other doubts that you have? Any other doubts?
Not as in the session. There are people better than me to clarify those doubts.
But any other doubts that you have in terms of your classwork? Anything bothering you?
>> Oh, what what what AC classrooms something that I should think about.
Yeah.
AC classroom.
What was that?
Oh, the charging points in the classes.
All right. Noted.
I' already checked AC.
So based on your feedback, this trimester has only got one mid exam.
Happy about it?
>> Yes sir.
>> Done with the exam?
>> Yes sir.
>> Got some good marks?
>> Yes sir.
>> Good. Are you all maintaining your CGPAs?
>> Yes sir.
>> That's important. That's not the only important thing. That's also one of the important things, right?
Any other difficulties? Any other concerns?
No. If you speak like that, I wouldn't understand attendance.
PBL classes without attendance, it doesn't work.
Even if we say 0% attendance, you still cannot sit at home foreign language.
You want extra classes for foreign language.
What was that?
Then I will understand.
O what's the problem with foreign language less time B tech four years less time no foreign language doesn't get completed by one single course. There is one more course also.
Immediately in the next semester you have one more foreign language course in the same language not not a new language. So that's a continuation.
All right. Let me tell you what is the philosophy behind having a foreign language course having it right from now and having two courses.
There are ample of opportunities in countries like Japan, Germany, Korea, especially for students who belong to circuit branches, right? CSE and allied, EC and allied branches.
So, we want to explore all those opportunities. You can only explore those opportunities if you're fluent in that language. No talking.
Hey You can only explore those opportunities if you know the language.
It's not sufficient if you just know the language. You have to be strong in the language.
So you get placements only when you are certified in the language. Just like how you have a certification for English lingua skill certification. Similar to that all these foreign languages have a certification.
All these foreign languages have a certification and they have multiple levels in that certification.
The goal is to at least crack the first level. Is that sufficient for all the jobs? No. It is only sufficient for some of the jobs. For those students who want to go ahead at a later stage, we will train you in the next level certifications at least with the current courses that is the foreign language course that you have in this trimester and the foreign language course, the second course that you have in the next trimester. Put together, we want you to achieve level one certification in that foreign language.
With that you will have clarity on how the language is, what the career opportunities are and you will then dive deeper.
Okay. And why did we start now? It's a language. You cannot start in third year and get placed in third year. Right?
Only if you start now, you will practice better and by the time you go to third year or fourth year, you'll have strong grip on the language. You'll also have time to complete the next level certifications in that language.
Right? And when you grab those opportunities, there are bigger opportunities, bigger opportunities like 40 LPA, 50 LPA opportunities.
So that's the reason why the foreign language course is mandated for entire BTech and for PBL batch we have two courses one after the other. The regular students only have one single course. I guess they study for an entire semester.
Since you're studying for a trimester, we made it as two courses because the duration would not be sufficient for you to finish off the certification if it is one single course. Is this clear everyone? Are the faculty teaching well?
>> The Japanese faculty. If there are any sections where you have a bit of a concern with the faculty, speak to us.
Speak to Krishna sir or speak to me. We can always carry forward that feedback and ensure that it is rectified.
All right. How are all the other courses going? Well, how is the web uh the full stack course?
Little difficult.
That is the toughest course in the entire trimester.
So, you need to put in more focus on that course. Spend more time. Your faculty are putting in lot of effort to develop videos, share them with you.
So I want you to go back home, rework on the project, learn the concept, get back to faculty and ensure that you get all your doubts clarified.
You need to put in a lot of effort for that course. Okay.
All right. If that is all then we'll wind up the session. Um it's wonderful to have you all and it's wonderful always to interact with you all. Make sure you finish the Corsera course, complete the projects and submit them.
Okay. All right. Good luck everyone. You can wind up.
相关推荐
OpenHuman VS Hermes AI: Who Wins?
JulianGoldieSEO
285 views•2026-05-29
BREAKING: Microsoft’s New Image Generating Model Beat Out GPT 1.5 and Nano Banana 2
aimmediahouse
122 views•2026-06-03
Long-Running Agents — Build an Agent That Never Forgets with Google ADK
suryakunju
142 views•2026-05-30
I Made the Same Anime Fight Scene in Every AI Video Generator
NobleGooseAnime
295 views•2026-05-30
Nvidia Bets Big On AI PCs | New Chip To Power Windows Laptops | Technology | AI Updates | N18S
cnnnews18
3K views•2026-06-01
I Tested NEW Opus 4.8 on Four Projects (Updated LLM Leaderboard)
AICodingDaily
298 views•2026-05-29
3D Platformer Update - NO CAPES
SolarLune
294 views•2026-05-30
AI Doesn't Create Bias — It Inherits It
UXEvolved
176 views•2026-06-01
