# Transcript

**Dr. Nagendra**: [01:03] Good morning, all faculty members of NIE. As you are aware, maybe that your HODs might have communicated to you. Now we have decided to bring AI into the curriculum. So this is a very much essential thing that what we are supposed to do. Because, like, AI integration in the curriculum is happening across the country. More premier institutions have already done it. So we are also integrating AI into the curriculum now. So in that connection, Mr. Srikanth, our director, he is taking the initiatives here.

**Dr. Nagendra**: [01:42] Today in this presentation, now we are having two speakers. One is Srikanth Nadhamuni and Mr. Anand. Now to introduce Mr. Anand, he is considered to be India's top 10 data scientists and AI expert, and he is also a TEDx speaker. And presently, like, he's also a visiting professor of IIT Madras. And he has spoken on various platforms on TEDx. He has made his presentations both in India and abroad. And he is an alumnus of IIM Bangalore, IIT Madras, and London Business School. This is a very brief introduction about him.

**Dr. Nagendra**: [02:23] And now coming to the second speaker, Mr. Srikanth Nadhamuni. He doesn't need any introduction because all of us know him. He is our Management Board Management Director. And he is a technological entrepreneur with close to four decades of experience. He was head of technology for Aadhaar, where he led creation of Aadhaar technology. And currently he is CEO of Khosla Labs. He is serving on Government of India active committee for AI Centre of Excellence in Education, Healthcare, and Agriculture. And he is a proud alumnus of National Institute of Engineering, and he graduated from National Institute of Engineering in the year 1984. Am I right, Mr. Srikanth?

**Srikanth**: [03:09] Yes. But NIE into 1984-85.

**Dr. Nagendra**: [03:14] 84-85. Now I request Mr. Srikanth to take it over.

**Srikanth**: [03:20] Okay, wonderful. Thanks, Anand, for joining from Singapore on a weekend. It's lovely to be with all the faculty members here at NIE. I was there in Mysore for a couple of weeks. I was working out of there. I just moved back to Bangalore. If I were in Mysore, I would have definitely enjoyed coming there and meeting all of you. But I think this is the next best thing. Zoom works really well. And Anand has joined us from Singapore as well.

**Srikanth**: [03:55] So we're going to take you through a presentation that we made to the management committee and to the deans and HODs about a week ago. And these are some broad ideas on the AI transformation that an engineering college like NIE can, you know, go through. And Anand actually is going to do a bunch of live analytics on various datasets, and some of them are from NIE itself. And it will be interesting to look at our own data. Of course, this is all completely, it has nothing to do with faculty names or student names or anything like that, it's completely anonymized data. So it should be interesting to see the trends and how NIE is doing on various areas.

**Srikanth**: [04:54] With that, let me just try to share my screen here. Give me one second. I have two monitors so just adjusting. Give me one second. Wonderful. And now I'm going to share my screen. Desktop 1. You should be able to see my screen now. Are you able to see it?

**Anand**: [05:41] Yes.

**Srikanth**: [05:44] Wonderful. Okay. So let's jump right into it. You know, many people wonder whether, you know, this whole AI topic is hype or is there reality behind it. So let's go a little deeper, and there's, you know, data to show that it's both. So I put a couple of slides just to go through a quick history of... if everybody can maybe mute yourselves, that'd be wonderful.

**Srikanth**: [06:23] A couple of slides on a brief sort of history of AI and where we are at today and how quickly things are happening. This was more meant for the management committee, but anyway, I'll just quickly go through it. You people probably all know this in great detail. This AI journey started way back in the 1950s with Alan Turing and an important conference at Dartmouth which was, you know, where this term artificial intelligence was coined. And that led to various things, but the initial phases was led by more sort of symbolic, logic-based rules, rules-based expert systems, and so on. And that is considered now as the AI winter, where not much progress was made. And the kind of, you know, ideas that came out initially of, you know, AGI and a computer that could do pretty much what humans can do and so on and so forth way back in the 50s, that did not materialize during that AI winter. And many people sort of gave up.

**Srikanth**: [07:38] I still remember when I moved to the US to do my master's, I remember this black gentleman was sitting next to me on the plane, he asked me, "What are you going to do? What are you going to study in the US?" And I still remember telling him, "I want to do artificial intelligence." But nevertheless, things did not pan out in AI the way people expected, and things slowed down and not too many people were focusing on it.

**Srikanth**: [08:06] And then there was this whole focus on neural networks and deep learning and so on, which suddenly revived a lot of interest because it started delivering results, right? And you started seeing things like Deep Blue, you know, where the IBM machine defeated Garry Kasparov. And of course later many things happened. Geoffrey Hinton from the UK moved over to I think Canada and was working out of Toronto. And he actually hit upon these deep learning methods where these so-called neural networks inspired from human brain cells were capable of, without feature engineering, able to do amazing things in classification problems and all kinds of what was earlier ML techniques. Suddenly LLMs were starting to do things. In fact, Anand can talk a lot more about it. He in fact teaches a course on LLMs and the use of tools and techniques for data analysis.

**Srikanth**: [09:20] So as we move along, **the big transformation came out of this paper in 2017, "Attention Is All You Need," which basically described a transformer architecture.** And the erstwhile, you know, language-related architectures, the RNNs were not doing so well. They couldn't keep context very long and they were incredibly slow because they were sequential, recurrent neural networks. **This transformer architecture really changed the face of AI, both from a language understanding, understanding the meaning so to speak, as well as the speed at which it could do it.** The architecture itself was inherently, you know, very parallelizable.

**Srikanth**: [10:13] So of course, since then we have seen, we have all witnessed, from 2017... you know, of course BERT came out of Google, GPT-3, you know, in 2022 ChatGPT came out. And the first time we all used it, we were like astounded. You know, **human beings are the only creatures with a neocortex capable of language, and suddenly there is this other thing, other device which could also speak and respond and talk to us in ways that completely flummoxed most people.** Right? How is it even possible? Because for millions of years nobody could do this excepting human beings. So I think it's a huge moment in human history, when machines are able to not only master language, but also reasoning, which came much more recently in 2025 with the reinforcement learning techniques. And so...

**Unsure**: [11:13] Sir, sorry to interrupt. Please. Is it maximum limit for the meeting is only 100?

**Srikanth**: [11:19] There's no limit. This is my own Zoom call.

**Unsure**: [11:24] Faculty are getting that meeting is at the host's allowed capacity.

**Srikanth**: [11:28] Oh, really? Yeah. How is that? Let me see.

**Unsure**: [11:34] Sir, you can create one more host, sir. One more fellow can be the host.

**Srikanth**: [11:41] Oh. How do I do that? What do you mean, just make another person a host? Okay, Likith, where are you? I'm just going to make you a host as well, if you can, if that helps. That is strange. Let me quickly ask ChatGPT what to do. Give me one second. Anand, could you also check?

**Anand**: [12:15] Just checking, yeah.

**Srikanth**: [12:18] What it could be. "My Zoom connection is limited to 100."

**Anand**: [12:41] Okay, mostly a host account is capped to 100. To fix it, you need to increase the meeting capacity of the host's Zoom license. Before the meeting starts... Zoom currently supports...

**Srikanth**: [12:51] Oh, but I thought I have a business license. Anand, do you have a Zoom license too? Or do a Zoom...

**Anand**: [13:05] The cleanest thing might be if you could create another meeting on Google Meet and share that in here, we can pass that to the other faculty. So that way you'll be streaming on two platforms. And I join that as well.

**Srikanth**: [13:18] Yes. Okay, let me do that. Or I'll tell you what, let me send you one meeting and that'll save you a bit of trouble and just... yeah, sending that over to you.

**Anand**: [13:33] Okay, on WhatsApp?

**Srikanth**: [13:35] I've sent it in the same chat window as Zoom.

**Anand**: [13:38] Oh, I see. Okay, let's go quick back up. Chats. Okay. Likith, can you share this with everybody?

**Likith**: [13:52] Sure, sure.

**Srikanth**: [13:55] Okay, and of course I'll click it so that... I have ask to join. Okay.

**Anand**: [14:09] Yeah, I'm just letting you all in.

**Srikanth**: [14:13] Okay, but do you have to do this for everybody?

**Anand**: [14:18] Yeah, I think so.

**Srikanth**: [14:26] Okay, I cut out the mic on the other one.

**Anand**: [14:28] No, no, then the others can't hear you.

**Srikanth**: [14:31] No, I cut out the mic on the Meet, but on Zoom I'm on. Can they all hear me?

**Anand**: [14:35] The those on Meet can't hear you.

**Srikanth**: [14:38] Oh, correct. So we probably want to mute on the speaker, yeah, correctly. Yes, yes. Okay, okay. Can everybody hear me both on Zoom and Meet?

**Anand**: [14:52] With an echo.

**Srikanth**: [14:54] So you probably still need to mute yourself. Balaji, Raju, can you hear me on Meet?

**Anand**: [15:00] Oh, of course my speaker is turned off! Yes! Okay, I'm guessing everybody can hear me.

**Srikanth**: [15:08] There's still an echo.

**Anand**: [15:11] Okay. All right. Well, my speaker is muted so I won't hear anything. If you want to get my attention, you know, either wave your hand or type on chat or something like that. Okay. So coming back to... oh, now I need to present view my entire screen. So first I will let me view back again. So many windows here.

**Srikanth**: [15:54] Okay, so I am going to share on Meet as well. Okay, so I'm back. And if you can't see the deck or hear me, please at least put it out on the chat. Okay. All good, Likith? Can everybody hear me?

**Likith**: [16:21] Yes.

**Srikanth**: [16:22] Yeah, yeah, please go ahead.

**Likith**: [16:23] I can't hear you. So if you can just articulate, all good?

**Likith**: [16:26] Yeah, yeah, yeah, please go ahead.

**Srikanth**: [16:27] Okay, perfect. Okay. So, so, as I was saying, you know, LLMs with the capability of reasoning has really taken things quite far. And we are now able to do various tasks including programming and so on and so forth. So let's see what kind of things the LLMs are capable of doing. Of course, LLMs now are not just dealing with text, they are multimodal. They deal with images, video, voice, you know, even singing, and so on and so forth. Right?

**Srikanth**: [17:08] So, firstly, you know, **the IIT Joint Entrance Exam is one of the hardest undergraduate exams in the world. And apparently, some of the models have been able to be on top.** Right? The All India Rank 1, right, it apparently beat the first ranker. I wrote an article about this some time ago on the net. And that's quite astounding, right? It's like one of the hardest feats. In the International Math Olympiad, solving, you know, out of the six questions, these are for, you know, like high schoolers and so on, but still, the Google DeepMind AlphaProof was able to solve four out of six problems, which put it almost at the gold level of International Math Olympiad. Really, really hard test, right?

**Srikanth**: [18:07] In medicine, in the United States, if you want to practice medicine, you need to pass the USMLE exams one, two, and three, and it looks like it has passed all the three exams and it's done quite well. In the Bar exams in the US, it looks like it's at the 90th percentile. And it went from the bottom 10 percentile marks to the top 10 in less than one year. Right?

**Srikanth**: [18:32] So these are just indications. Many, many, many more things we are reading on a daily basis of what these LLMs are capable of doing. And **especially in programming, they are doing fantastic work.** You know, in some of the companies where I am, you know, that I am part of, we are seeing enormous productivity gains on programming, right? And they are able to do enterprise-class programming, not just writing small functions and so on and so forth. We are able to give it really complex tasks in very large, you know, code bases and they are able to do some amazing things.

**Srikanth**: [19:11] So, you know, all these kinds of new technologies, whether it was Internet, cloud, crypto, whatever, whatever the new technologies that keep coming, it always goes through this, you know, what is called the hype cycle. And Gartner's hype cycle for 2025 for AI shows this, right? And what's interesting is there's initially a trigger. In this case, deep learning, you know, the paper that came out in 2017 and eventually all these amazing applications of the transformer architecture, that was the trigger of this new AI wave. And immediately there's a peak of inflated expectations. A lot of hype. A lot of cheerleading. Lots of money coming in. VC money. Every company says I'm doing AI, every... you know, everybody says, you know, I'm already on AI and I'm the expert in AI and I've jumped into AI and I've created AI and so on. Right? So this is a peak of inflated expectations and hype that goes on.

**Srikanth**: [20:23] And then, when they realize, oh, we can't do all these things that are happily, you know, AGI solving world hunger every, you know, didn't happen. Suddenly there is a trough of disillusionment, right? Things like, oh, this doesn't work, right? And eventually it starts stabilizing. A slope of enlightenment where things start actually working because you've solved some of the issues around the new technology. And then you reach some sort of a plateau of productivity where things start working. Right?

**Srikanth**: [20:58] So you see this hype cycle work. So AI is no different. It's happening in the Gartner hype cycle. And you can see various pieces where they are at and so on. It doesn't really matter, you know, I don't go into detail out here, but point I'd like to make is, certainly there's hype. Right now there's a lot of hype meaning very inflated valuations of companies. Trillion dollars and so on and so forth. OpenAI just raised 105 billion dollars... oh, sorry. Which has raised that much money on a valuation of, you know, close to 150 billion dollars, which is unheard of for a new company, you know, a 10-year-old company. Right?

**Srikanth**: [21:36] But... so that definitely shows the hype. And people talking about GDP growth is 20%, 30%, and so on, which are sort of hard to believe. So there is certainly hype. But there is also on the right side, there is product, a plateau of productivity. We are already seeing in many areas. Okay.

**Srikanth**: [21:57] And so what does all this mean to us at NIE? Okay. So well, even before we get to engineering, what is the impact on the labor market? Because everybody is concerned, right? Whether it is professors or researchers or people in the, you know, programmers and so on and so forth, a lot of people in the industry. And Anthropic, one of the leading labs that has put out the Claude model, a very, very successful model, LLM, has just last week put out this paper from their own data of people using Claude, they showed which sectors are at most risk, you know, of losing jobs and so on. And they highlighted Management, Business and Finance, Computer and Math. Right? A lot of the engineering, computing architecture, engineering. Life Sciences. Social Sciences. Legal. Arts and Media. Office and Administrative support. Okay.

**Srikanth**: [23:01] Of course, things that require a lot of manual work, Agriculture, Construction, Installation and repair, Production, Transportation, they are very... they are not impacted. They are impacted very little as you can see, right? So this is the analysis that just came out about a week ago based on actual data that Anthropic has seen from their clients. Right?

**Srikanth**: [23:24] And what does all this mean for NIE? And for the faculty? And for the future of engineering education? This is not specific to NIE. Okay, this is not even specific to, you know, Mysore or whatever. India, globally, everybody is trying to figure out how do we sort of deal with this new AI transformation and what does it mean.

**Srikanth**: [23:54] Qualities our students need to come out with, right, when they leave our college, graduate from a college, things have changed drastically. Okay. Like I already said, AI can write code, it can write... Hey Anand, did I skip... oh sorry, I'm so sorry. I forgot to get you in on the other slides. You want to jump in now? Where do you want to...

**Anand**: [24:18] No, no, go ahead.

**Srikanth**: [24:20] Say that again?

**Anand**: [24:22] No, go ahead. We'll switch to NIE specific datasets at the end. We don't need to do this.

**Srikanth**: [24:28] Okay, so I'll finish the first set like what I had in the slides before getting into the curriculum. And why don't you... do you want to do at that point, you switch over to your data analytics?

**Anand**: [24:38] Yes.

**Srikanth**: [24:39] Perfect. Perfect. Perfect. Okay, so we can already see that AI can write code, it can generate reports, it can generate documents. In fact, a lot of this deck Anand and I created using AI. It's quite incredible. You should now, you'll see him, you know, do the analytics on some very interesting data pretty soon using AI.

**Srikanth**: [25:13] It can... it can do various things, as you can see, it can, you know, read images, diagrams, speech, data, and so on. And we already went through what is under threat, you know, the things that use computers, the white-collar jobs are certainly going to get affected, right?

**Srikanth**: [25:34] **What becomes valuable?** Okay, this is a good question to ask. What do we teach our students? And even as a faculty itself, what is going to become more valuable? Okay. **Strong engineering fundamentals.** Okay. **Problem solving, problem framing, and systems thinking. Judgement and trade-off analysis. Learning how to use LLMs.** In fact, Anand is going to, he offers this course where he teaches you how to write prompts, how to get the best out of LLMs. Right? So we'll see some of that. Those become really important. Right?

**Srikanth**: [26:08] Now, on the second row, the first point is, **it impacts all branches. Not just computer science.** Okay, every branch of Science and Engineering, STEM education is changing because of AI. Right? What must colleges do? We need to start teaching our students AI as foundational courses. We have to embed AI in every discipline. Okay, we'll go through that. There's another slide on how, what we need to actually do.

**Srikanth**: [26:42] And, you know, the new graduate that comes out is not just someone who knows Java or Python or syntax of coding, okay? **They need to become AI-native engineers.** Okay, and we'll see a glimpse of what that is, right? Not just a user of AI tools, they need to be able to design. Right? In our companies, people who were developers are no more just doing development. They've sort of moved up one level to become product creators. They not only design products, but they also develop products. Because just development of products is not as useful. Claude Code does it amazingly well already, okay? And people are not hiring as many junior engineers. So the big question comes, you know, if we, you know, if the big attraction to colleges like NIE is getting a degree so that you can get a job, and many of them are IT jobs and so on, which is where we have been seeing a large amount of demand and increased number of seats and so on and so forth. If those youngsters don't get a job because AI is going to code, then what happens? Okay, what changes do we need to make within NIE to ensure that we can accommodate for this new world? That's what we're talking about.

**Srikanth**: [28:05] Okay. So before I move on to this, Anand, why don't you take over and go over your ideas?

**Anand**: [28:15] Sure. Let me share my screen on both. I think you'll need to stop sharing.

**Srikanth**: [28:23] Okay. Sharing on Zoom to begin with, and it should be visible any second. Next, you'll probably need to stop sharing on Meet as well.

**Anand**: [28:41] Oh, yes, sorry. Good point. Stop sharing. Yeah. Great.

**Anand**: [28:53] Okay. Let me give you a history of the rough capability of different models. See this is a chart where each dot is an LLM. There was a time when, around 2023, there were really only three popular models. And out here, the x-axis tells you what is the cost of these models. And that can range from Claude taking $8 per million tokens. One million tokens is roughly how much it costs for it to read the entire King James Bible or the entire seven-book Harry Potter series. That volume is one million tokens. That for Claude cost about $8. GPT-3.5 Turbo was also one of the leading models at around that time and it cost only 50 cents. Literally one hundredth, oh sorry, one tenth, less than one tenth of that cost.

---

**Anand**: [00:00] The models evolved in both cost as well as in intelligence. The intelligence is on the vertical axis. These models were somewhere between a high school class 6 and class 10 student. High school graduate versus high school freshman. So maybe a class 8 level intelligence is what we had in 2023. Fast forward a little bit, and in November 2023, when GPT-4 was introduced, we had, for just about the same price as Claude 1, around $10, we could hire a college junior. Someone who has just entered and will be able to submit assignments, do calculations, etc., at roughly that level of intelligence.

**Anand**: [00:58] Now the interesting thing is, just about eight months later, that same cost fell to 15 cents. $10 down to 15 cents. That's 60 times cheaper. So, **instead of one college graduate, for the same budget, I can hire 60 college graduates.** That proceeded, and with GPT-4.5 preview in Feb 2025, we had Ph.D. candidate level intelligence as the first model that came up at around Ph.D. level intelligence. Today we are at not just Ph.D. level, but with GPT-4.5 Pro we reached a tenured professor's level of intelligence approximately. And now are beyond that with models like Claude 4.6 Opus. And they keep becoming faster and faster. If I want to hire...

**Srikanth**: [01:48] Anand, I don't think your audio is... I am hearing your audio. It looks like some of them are saying they can't hear the audio.

**Anand**: [01:56] Oh, on Meet they wouldn't be able to because... okay, yeah fine. Now it's going to be a little tricky. Okay, I'll see how we can minimize echo. Now the others should be able to hear me though.

**Srikanth**: [02:18] Anand, you might want to start from the beginning on this slide, maybe I don't think they heard you.

**Anand**: [02:24] Got it. I'll summarize. Though if you unmute, Srikanth, I'm getting... okay, yeah, now the echo is gone. Thanks. Let's quickly run through this chart. Each dot is a model. The X-axis is the cost of running the model. For example, GPT-4 costs $30 to process the entire Harry Potter book series just to read the whole thing. Whereas we have smaller models like Gemma 3, which are actually smarter and can process the same amount for two cents. Look at the difference. $30, one-tenth is $3, one-tenth of that is 30 cents, one-tenth of that is three cents. So, **Gemma 3 is less than 1,000 times the cost for superior intelligence.**

**Anand**: [03:15] And it was released when? Gemma 3 was released in May 2025. GPT-4 was released in March 2024. So, **just 14 months later, the cost fell by 1,000 times.** And that's the speed at which the cost is falling. The intelligence now that we have is roughly the level of a tenured professor's level of intelligence.

**Anand**: [03:45] Now, imagine this. If you give a professor a book roughly the size of Harry Potter and said, do all the analysis, come back to me. And they come back for $5, that's what, 400, 500 rupees? And they say, this is the kind of analysis... and it's a very comprehensive analysis, and it takes maybe two minutes or so. That is crazy. **This is the kind of analysis that a McKinsey consultant would charge several, possibly hundreds of thousands of dollars for.**

**Anand**: [04:26] So I said, why don't we then take NIE as a starting point. If NIE hired a McKinsey or a BCG to do a consulting engagement for them. The consulting engagement has the following objective: **How can we improve our NIRF rankings?** So the first question that I asked this consultant is, my aim is to help the NIE Mysore faculty understand how AI can help transform their work and the institution. And one lens is the NIRF ranking improvement. What factors can improve the NIRF ranking and how can these help with AI?

**Anand**: [05:01] You go ahead, do the research and figure it out. Now this was on Sonnet 4.6 with extended thinking. Sonnet, in terms of its capability, is somewhere at a Ph.D. level. Yeah, it's at a Ph.D. candidate level and slightly cheaper than the Opus model. Why didn't I use the Opus model? Because I didn't want to run out of my $20 credit, and Sonnet is 95% there. Good enough. Why extended thinking? Because that will nudge it to go deeper, probe more, and so on. Could you have done this with ChatGPT? Yes, GPT-5.4 is almost at the same level at the moment, but yeah, it's probably very close to the same level. Gemini 3 is very close to the same level. So it doesn't really matter which of these you use, but if you are using the paid models. If you are using the free ones, these don't quite reach that level of quality.

**Anand**: [05:51] So pick any $20 or even the $400 Pro subscription for ChatGPT is also fine. And you get in your pocket a practically free McKinsey consultant who can answer dozens of questions like these for you. And here is the response. There are five parameters that matter: the TLR 30% weightage, Research and Professional Practice 30%, Graduate Outcomes, Outreach and Inclusivity, Peer Perception. And of course, the first two are clearly getting much higher weightage, and we can look at the subcomponents.

**Anand**: [06:31] So when it comes to teaching and learning resources, the biggest weightage is for faculty-student ratio with permanent faculty. For Research and Professional Practice, **the two biggest chunks are publication volume and citation quality.** That's what you need to focus on. Now, where are we as NIE Mysore today? Starting from 2018, we have steadily declined in rank. Starting from a rank of around 100, we were at in 2022 a rank of around 200. It jumped up dramatically to around 151 but stayed around there. And if this curve had continued, we could have reached by now the 100 rank that we held not too long ago, less than a decade ago.

**Anand**: [07:28] How does this break up? The biggest gap is Research Practice, and PR, Peer Perception as well, which is relatively low. The strength is in Graduate Outcomes. While teaching and resources are not too bad, outreach and inclusivity are not too bad. And therefore, from an intervention perspective, it's advised in terms of AI levers is AI-assisted lecture preparation and adaptive content. As the world changes, you modify your content accordingly. This has high impact. I'm not even going to go through the medium and low ones.

**Anand**: [08:18] But even more critical is in the RPC side. Can we do literature review and gap identification? What is the point of publishing in an area which is already overpublished? **Identify the underpublished niches where you can start building your citation clusters directly.** Second, grant writing. This is a pain. So what if we identified what applications to write, which organizations to pitch this to, and move towards a significantly higher RPC score? Again, very high priority. And there are a whole series of things that one could do.

**Anand**: [08:56] So based on this, the next question to this consultant is, we have the publication status tracker. Some of you have filled this in. In any case, the information is out there publicly. We know exactly which department, which author is publishing which paper. And we also now have the details on which ones of these are in a draft stage. That is useful information. The output of any system is as good as the input we feed it in. And this is rich information.

**Anand**: [09:30] And for you to get a quick sense of what that publication tracker is looking like, this is the information as of this morning. I had fed it in a few hours ago. So the last maybe 30, 40 rows are not yet fed in. So I will be, and maybe not I will, you should be taking the revised data and sending it back with similar prompts and asking it what it thinks are potential areas for intervention.

**Anand**: [09:56] But based on what we have as of this morning, I said, how can we do AI-accelerated literature review, grant writing, and give me specific, actionable, high-impact, low-effort recommendations? I don't want to waste time doing a lot of work. The effort has to pay off quickly. So it spent a fair bit of time, and the amount of thinking that it does is not small. It goes on and on and on and on. And eventually, it says, let's start with literature review and gap identification.

**Anand**: [10:29] Step one, here's something that we can do immediately. There are or were this morning about 26 papers in the draft stage. Now I'm sure there are more. There were 14 in ISC, 3 in ECE, 2 in CSE. And what each of the authors of the papers could do is, number one, go to Elicit.com. It's an AI scholarly search engine. And ask, what are the top 20 cited papers on [put in your topic] in the last three years? Some of you may have already done this, which is great, and if you haven't, then do it right away.

**Anand**: [11:15] For each of these, specifically in ISC, ML, healthcare, ask what are the open research gaps in, pick your topic, prenatal care prediction, blood pressure biomarker discovery, etc., using machine learning as of 2024? And that starts with a gap identification as well. Put it directly into Claude. Give your draft, the literature review draft, saying, "I am writing a paper on this particular topic. Here are the gaps from my literature review. Which gap does my proposed method best address? And rewrite my introduction to foreground that particular gap."

**Anand**: [11:46] What this does is surfaces the relevance of the paper significantly more. And the reviewers are not only going to understand the impact of this, you will get an insight into how you might want to tweak the draft as it progresses to specifically addressing a known gap in the literature. Which is obviously helpful.

**Anand**: [12:08] Second thing, Physics department. Let's take the AI for crystallography paper positioning. Go to Semantic Scholar. This is pretty good even compared to Google Scholar, though there are several other search engines. You can use typeset.ai, a bunch of others. And ask for the thiazole crystal structure, DFT molecular structure docking. Get the top 50 citations, put it into Claude and say, look, here are the 50 recent papers in my field. And it will go through them exhaustively. Identify which of these subtopics are overpublished. What combinations are appearing rarely? And therefore, **suggest three underexplored angles for my next paper.**

**Anand**: [12:53] Again, the theme is, use it to find out what research is already happening. Stay clear of that area. Use it to find what research is not yet happening. Show in your paper how you are using or filling that gap, and see, also use that to deepen the research in that particular topic. And these are easy. This in fact could be done right now in this call. And maybe we should. But what I'll do is go breadth first, and then we can deepen. But I would encourage those of you here to try this out now. And it doesn't have to be Claude. Claude is saying use Claude. Use Gemini, use ChatGPT, any paid subscription that you have. Do it now, and I'd love to hear from you what you found that was useful and not useful. Because that's how, by discussing, we will learn how to use AI better.

**Anand**: [13:45] Near term, for ECE and ISE, let's build department citation cluster maps. So for instance, Ramya who has the highest output overall with 18 papers as of this morning's data, Vijay Rajalakshmi, Srikanth, Mahesh, all are independently building citation volume. What if we started coordinating the topics into a citation cluster? Now that can significantly improve the top 25 percentile score metric.

**Anand**: [14:15] Now how might we do this? Let's go to ConnectedPapers.com. Again, it has a free tier. And what we could do is put in the most cited Scopus paper that we have, download the graph, and feed that into the AI. Here's the citation graph for memristor-based hearing aid research, blah blah blah, and put in all the paper titles and their citation counts. Find the three papers that would most benefit from a survey or review article citing them. **What are the co-authorship opportunities that would create Q1 level citations?** And which of these would map to a funding scheme as well?

**Anand**: [14:56] So you're hitting three birds in one stone. One, you have the benefit of getting into a stronger citation cluster. Second, you are identifying who else you can co-authorship next, which will have the potential for improving the strength of that co-authorship cluster. Third, you're also exploring where this could lead to funding. Now we know who needs to take that action. We know what action needs to be taken. And it goes on like this. I'm not going to go into each of these specific recommendations and there are a series of things that you could do on the medium term as well. And department-level playbooks. For each of these departments, what could we do? Let's take for instance Physics. Let's from a literature review focus perspective, explore thiazole and triazole drug discovery gaps, and so on.

**Anand**: [15:53] Oh yeah, I completely forgot grant writing. Let's come to grant writing first. Here are the best fit grant schemes based on your research. If you wanted to go to the industry, for instance, SAP Labs, Cisco, Mercedes-Benz, they're already recruiting from NIE. They have a CSR to lab route. So, **taking CSE and ISE work, pitch to SAP on enterprise AI. This takes practically zero incremental effort.** Taking the ECE work, pitch to Cisco on networking and IoT. Mechanical work, pitch to Mercedes on EV manufacturing. Now you may be doing this. Maybe you're doing 80% of it, 20% of it, but wherever there is a gap, that is the starting point.

**Anand**: [16:33] And the next question will be to say, oh, but out of these three that you gave me as recommendations, two of these I'm already doing. Tell me more. Which it does. AICTE, what might you want to pitch there? SERB, what might you want to pitch there? MeitY, what might you want to pitch there? And another next step is, now write me the grant proposal in a way that will maximize the likelihood of conversion. Better yet, research what grant proposals have been made. Even better, research who are the people that are likely to be reviewing the grant proposals. Tell me what it is that they are specifically looking for based on their recent grant awards as well as recent grant feedback and topics of interest. Based on that, write me a grant proposal that will go through this particular person's hearing.

**Anand**: [17:31] **You effectively have the equivalent of a tenured professor sitting in your pocket for $20.** Worth exploring. This is one kind of analysis that you don't have to do much on. You just have to feed the data into Claude, ChatGPT, Gemini, whatever. And I'll come to this bit later, and work on this. This was one data set.

**Anand**: [17:56] Let me move to the next data set. But I will pause here. I know I've been speaking for a fair bit of time and you've heard Srikanth extensively cover a lot as well. You're speaking but we can't hear you, Srikanth.

**Srikanth**: [18:18] Your echo has increased. Wondering if something, some setting changed.

**Anand**: [18:24] Not sure. But is it still echoing now? It's not echoing for me.

**Srikanth**: [18:30] Yes, there's quite a bit of echo. Yeah. Likhith, for you also, you are hearing the echo?

**Likhith**: [18:44] Yeah. From you.

**Anand**: [18:47] Okay. Yes. Is it still echoing for you, Srikanth? Maybe just do a thumbs up, thumbs down. Don't unmute yourself please.

**Srikanth**: [18:56] No Anand, it's fine.

**Anand**: [18:58] Okay, you're fine. Thanks. Thanks Likhith. Great. Srikanth, there is an echo from your system, so...

**Srikanth**: [19:04] Well, still echoing for me, as long as it's fine for everybody I'm fine, but...

**Anand**: [19:10] My suggestion is, right-click on the Google Meet tab on top and mute the tab, Srikanth. That will give you only one source. For anyone who's joined in on both Zoom and Google Meet, you will hear an echo, because I'm speaking on both.

**Anand**: [19:30] Okay, but if there are any questions at this point, please don't hesitate to put this in the chat window, either on Zoom or Meet, and we can take that as we go along. I will now move to the next data set.

**Anand**: [19:48] We have student feedback data. This data is effectively across three cohorts. And now I had coding agents work on this. Many of you are likely familiar with coding agents like Codex, Claude Code, etc. These are effectively the same as ChatGPT or Gemini or Claude, with the difference that they are able to write and run code reasonably reliably. Now ChatGPT, Claude, etc., can also write and run code. These are running on dedicated machines. You'll say, wait, hold on, they also run on dedicated machines. Yeah, true. The difference between these is a little hazy, but the coding agents are a little more focused on coding. And sometimes I tend to use them when I want to do more deeper data analysis, especially with data that is sitting on my system, rather than having to upload everything.

**Anand**: [20:45] So, here's what I did. I asked Codex...

**Srikanth**: [20:53] Anand, sadly people on Google Meet are unable to hear you.

**Anand**: [20:58] Okay. Interesting. Let me try something. Are people on Google Meet able to hear me now? If someone could put on the chat window a yes or a no, that will help. Okay, that's a thumbs up. Great. I could see that. Thank you. I'm guessing the Google Meet solved it. Yes. Thanks everyone. I can see that.

**Anand**: [21:24] So what I did was told Codex, and then Claude Code, to analyze and then visualize the student feedback data. I'll show you the results, then we'll come to the prompts. Here is the result. It built a full-fledged website, effectively a mini presentation of sorts, with what we are learning from all of the student feedback. That **the teacher in the room is everything.** Let's drill down.

**Anand**: [21:56] Say you are a sixth-semester student in ISE. A top company is visiting campus tomorrow, but you have 84% attendance. One more absence and you lose your exam eligibility. You sit out the interview. You miss the whole thing. And this is not a rare story, according to the feedback. 2,500 students filled out their teaching feedback forms. They weren't just rating the professors. They wrote about broken lab computers, they wrote about instructors who were reading from the slides, they wrote about the 85% attendance rule, **systems that seem to be protecting the institution, not the students.** And in this, there are a whole series of findings that need some serious attention, potentially even action. Let's go through those.

**Anand**: [22:45] Firstly, what do we have? We have data from 735 in the first year, 524 in the second year, and a general cohort dataset of 1277 students. Several suggestions coming in as text. And by department, this is the mix of responses. Here's what we are finding. First years versus third years versus fourth to sixth years. First year student ratings tend to be clearly higher. Third versus fourth to sixth, there isn't that much of a difference, but the dip is happening here. The steepest fall is happening between the first year and the senior cohorts. That's where the disillusion is happening. And let's talk about why this is happening also. I mean, the hint is already here. Lab quality seems to be a primary determinant. And which function will also become clear in a short while, but let's proceed.

**Anand**: [23:34] You'll notice that the gap between the first year score and the fourth and sixth semester scores... the pink is the later student scores. Let's take on the right side, faculty accessibility. Not much of a difference. They don't have a problem there. Of course, they do have a problem, but it's not worsening. Course structure is also not worsening. Faculty preparation maybe marginally. Lecture pace maybe marginally. **But what is clearly worsening in their opinion is lab quality and lab manuals clarity.** Also, real-world examples, but going back, lab reinforcement of theory. So the labs are where between the first years to subsequent sems, there seems to be a gap.

**Anand**: [24:21] What predicts the overall satisfaction? Short answer is, faculty preparation. What do you mean? So let's take the correlation of the overall learning versus the individual parameters. What we are seeing is that **the correlation for faculty preparation is the only one that's seriously strong.** Significantly high correlation value, 71% correlation. And the rest are... I mean, lab quality, we already saw, was a pretty important one. But what exactly do we mean when it comes to faculty preparation? That will become evident in a short while. We'll go in there.

**Anand**: [25:04] But let's also quickly look at the department-wise feedback. This is something very likely that you would have seen. Which is that IP, with a relatively small batch of 30, is doing relatively well. Among the larger batches, Civil is close to 4. ISC is close to 3.3, and in that, there is almost a 0.7 gap or 0.6 gap in terms of ratings. The dimension that seems to be weakest across almost all of these incidentally is faculty accessibility outside of class. Let's take Mechanical Engineering as an example. This is the mix by cohort of the different parameters, and you can see that the consistency of accessibility being relatively low is out there. Lab quality mixed. Initially they thought it was pretty good, and then they thought not so good. This is the general category.

**Anand**: [26:00] We have pages and pages of qualitative feedback, completely unstructured. The coding agents were able to extract recurring themes. Lab infrastructure is a huge problem. Outdated PCs, MATLAB crashing, mechanical labs lacking working equipment, 85% attendance mandates causing missed placements and internships. Now, obviously, not all of these represent a reality that must exist in exactly the same way, but this is a fairly widely represented voice in any case. And what does that specifically mean? I'll come to the quotes in a short while.

**Anand**: [26:07] But there are a series of recommended actions that you can explore. For example, can we upgrade the computer hardware and stabilize the internet connectivity in the CAED and computer labs specifically, because that's causing a decent amount of technical downtime. Or, can we standardize the distribution of study materials and module summaries immediately after each module is completed? These are what AI is literally made for. Record the session, upload it, tell it to create the study material, upload it, and the students will have it. In fact, for this session, that's what I'm going to do. We are conducting a session, and I'm going to create concise study material and module summaries immediately after this, and I'll show you how. I've done it in the past as well, and upload it. Problem solved. Now, not all of these are AI solvable, but many of these were AI identified in the first place, which is of course also helping.

**Anand**: [27:08] What are the signals that we are getting? Let's take pedagogical delivery and pacing. Students are saying go too fast to complete their portion. That's one feedback that has a reasonably strong cluster. Rather than covering the syllabus, teach it. Stop reading from PowerPoint presentations and explain the concept. **Board teaching is more effective, please use it.** So could we shift maybe from static presentations to interactive board work? And interactive board work does not necessarily mean us drawing. It could be typing, building stuff, and AI can be a powerful lever here. What I'm doing here right now, for instance, is an example of material that was prepared this morning and is being presented right now. And with AI, you will be able to customize this material much, much better than I do in any case. More importantly, with the data, you'll be able to query it yourself. And that allows us to modify our teaching styles to something that leverages AI. I don't have to know all the answers. I just have to make sure that you are able to learn what you need to learn.

**Anand**: [28:18] It goes on, but where I'm going with this is that, number one, it's a reasonably large data set, it's a reasonably unstructured data set, and we are able to analyze this in a duration of approximately... **this entire analysis took half an hour with me not doing the analysis. I was doing yoga in the morning. While I was doing yoga, this was doing the analysis.** And now...

**Srikanth**: [28:44] Anand, can you show us your prompt? What prompt led you to get all of this?

**Anand**: [28:50] Absolutely. So let me show you the prompt. It'll take approximately 10 seconds. Let's specifically take the student feedback prompt. Make it a little bigger, and yeah, that should be clearly legible. You are a data analyst agent. Your task is to analyze the student teaching feedback from NIE, produce a department-level insight for faculty and leadership to act on. Here is the content. And all three files have the same schema and one open text suggestion column. I want you to merge the files, normalize the column names, compute per department, per cohort... blah blah blah, for each of these.

---

**Anand**: [00:00] Cohort, blah, blah, blah, for each of these collect the nominal text, do a cohort comparison. And at this point, you're saying, "Anand, how did you know to do all this?" And the short answer is I didn't do all this. This prompt itself came from Claude. So how did I prompt to get to the prompt? That's what I should show you next.

**Anand**: [00:20] What I did was, you saw that earlier analysis on how we can improve the NIRF, right? So apart from sharing with it the publication tracker, I also shared the files that I got. And I said, not the file content itself, but an analysis of the files. And I shared the following content with me. This is the inventory of all of the files that were shared. What can I do that will showcase the power of AI as part of their work and can improve the NIRF ranking? Give me multiple analyses that I can do and evaluate them on the two criteria: where does it show the power of AI, and has high potential. And give me the three most promising ones to execute.

**Anand**: [01:03] It did a churn and it said, look, here are about a dozen analyses that you can do. X-axis is NIRF impact. Number two is highest on that. AI showcasing also number two is highest on that. Idea number one and idea number three are also very good. Here are the ideas. Idea number two is for instance, identify structural versus random shortage patterns. That is, are there courses where the shortage rate is consistently high across months? So what that will tell us is that it's a teaching delivery problem, not a student problem. And it gives me some sense of how to do all of this.

**Anand**: [01:37] I said, boss, this is all fine. I don't want to do all this analysis by myself or even figure out how to do it. You've done the thinking. Give me the prompts for the top three so that I can just give it to a coding agent. And it gave me the prompts. I just copied the prompt, and that is what you saw. Yeah, "You are a data analysis agent, your task is blah, blah, blah." I just copied this prompt and put it in here.

**Anand**: [02:04] **This is meta-prompting. When I don't know in detail even what to ask for, I ask it.** But there are some parts where I know in detail what I want. I don't know enough about NIE, but I know enough about data visualization. That is my field. So there, I gave it my prompt, which is: I want you to create an index.html, data story. I want you to specifically use a data story skill, which is a collection of all of my expertise in building data stories. It's a separate file there. I want you to use tooltips for this particular purpose. I want you to use pop-ups for this particular purpose. I want you to use animated SVGs for this. All of this is stuff that I have learned over the years, and there I don't need meta-prompting. There I will directly prompt. **If you know, prompt. If you don't know, meta-prompt.** In other words, ask it for how to prompt, and then get the job done.

**Srikanth**: [02:54] Anand, quick follow-up question. The data visualization expertise that you have, and you're fairly unique in this, that portion of it, is it pretty much cut-and-pastable so that you can focus on the domain, and this visualization happens automatically? Or do you need to know quite a bit about visualization?

**Anand**: [03:16] It's cut-and-pastable. A lot of expertise is cut-and-pastable, and these are becoming popular by the term skill.md, or Claude skills, or Anthropic skills, and so on. My expertise, so to speak, is condensed into one data story skill—actually two of them. This is how I write narrative data stories. It says, write like Malcolm Gladwell, visualize like the New York Times graphics team, etc.

**Anand**: [03:45] Now, where my expertise comes in handy is I look at the output and say, no, no, no, change this, change this, change this, etc. But there are so many areas where I have no clue. I know nothing about NIRF, where my expertise doesn't even matter. **I will just take its expertise, and it is anyway at a tenured PhD level and above.**

**Anand**: [04:08] With that, I'll wrap up this part, hand over back to you, Srikanth, and maybe stop sharing screens as well.

**Srikanth**: [04:18] Let me restart Meet because of the problem I was having on this feedback. I had... let me get the Meet link once again and restart Meet so that the Meet group can also see my presentation and hear me. Okay, I've joined now. I'm going to share... Anand, is your analysis on that student feedback completed?

**Anand**: [04:58] Yes.

**Srikanth**: [05:00] Let's just... presentation. So much going on here. Okay, there it is. Okay. Are you able to see my deck?

**Anand**: [05:27] No, we see the other screen, Google Meet.

**Srikanth**: [05:31] Are you able to see my deck?

**Anand**: [05:32] Not yet. We see Google Meet.

**Srikanth**: [05:42] Yes. Okay, there it is. Okay, all of you should be able to see it now. Can you all see it? Both Meet and Zoom?

**Anand**: [06:28] Yes.

**Srikanth**: [06:30] Okay, perfect. Okay, so now that was a precursor. Three big ideas that we have for NIE as to with the AI transformation that has been happening. Oh, not audible? Okay, yeah, that's because... okay, let me turn on one of them. Oh, actually I have to turn on both of them. Turn off my... Okay. So... yeah. It should be fine now. Is it fine now? Can you send out a few thumbs ups? Okay, wonderful. Beautiful. Sharath, thanks. Raju.

**Srikanth**: [07:18] All right. So now with that as the background, let's talk about what we need to do. We saw what's happening with the AI. We saw what AI is capable of doing. Anand took us through quite an amazing demonstration of what NIE's data is showing, both from a research perspective as well as student feedback perspective. And by the way, we can institutionalize all of these things that Anand is showing us.

**Srikanth**: [07:50] Now, what is the big transformation all about? Three big ideas in my mind. Okay, so the first is an AI-integrated curriculum. Like we said, every branch is going to change based on AI. There will be AI itself integrated into the branch. There will be AI that we can use to deliver lessons particular to that branch as well. Of course, the fundamentals of AI perhaps are mathematics and computer science. Things like linear algebra and probability and statistics and perhaps discrete math and so on. And perhaps algorithms, some CS concepts. But the effect of AI is everywhere, and all branches have to participate in this, including our core math and the sciences as well, right? Now that's the first idea, and we'll go into what does this new AI-integrated curriculum look like.

**Srikanth**: [09:00] The second part is this: how are we going to transform this college into this AI curriculum? So here we already have some help. There is fantastic courses out there, digital content that's available. So the idea is, can we merge these two ideas of digital content and physical classroom and deliver this new AI-integrated curriculum better? That is the second idea.

**Srikanth**: [09:30] The third idea is **we need to move away from theory-based, memory, rote-based learning techniques that have been there for decades. It's not just an NIE problem, it is across the board. Can we move more to practice, hands-on learning?** So the idea is if we can do one, two, and three, we feel our students can have a much better chance of actually contributing enormously to industry, to research, and so on, to learning.

**Srikanth**: [10:08] Okay, with that, let's go into these ideas. Okay, so with respect to the new sort of AI-integrated curriculum, what we thought was there's a whole bunch of things, for instance, Anand has taken us through, the various LLMs at different levels of sort of capability, their costs, and what they are able to do. He's also shown us how to meta-prompt and get it to generate prompts that can be used by other AI tools to do our work, whether it is in research, whether it is in looking at student feedback, and so on and so forth. Right? Getting a coding agent to code and create a website of what our students are saying all in 30 minutes. This is enormous capability. But how did he get to where he is?

**Srikanth**: [11:02] Can we all learn that? Can we get that foundational element, whether we are civil engineers or mechanical engineers or computer scientists or mathematicians? Can we learn to do all these pieces so that we can get much better at not only teaching it but also in our own subjects, right? Can we go deeper? Can we understand what's happening out there? Can we contribute to that research in a much better way, and so on.

**Srikanth**: [11:27] So there needs to be some foundational element of learning about AI literacy for the faculty as well as our students, which brings them up to a certain level. This is like the basic. You need to know whatever Anand was showing today: the prompting methods, using visualizations, understanding data, and so on. All of us need to know this. Okay, so this we are calling the Common AI Foundation. On top of this we can build expertise in the various other fields.

**Srikanth**: [12:00] So AI literacy for engineers, what can LLMs, machine learning techniques do? Capabilities, failure modes, hallucinations, prompt design, and so on. Now there is also data and experimentation. Anand went through quite a bit of this today. He looked at our own data and showed us, you know, what can we do, how can we analyze this data, how can we visualize this data, where the decisions are jumping out of the graphs. It's not hard to figure out what's going on. It's not a complex web of, you know, we have tossed all kinds of data and we don't know what to do with it, right? It was very clear that, hey, we need to move in certain directions. We need to put our efforts in certain areas. And so on. So there is this part of it.

**Srikanth**: [12:48] Then there is the department-specific curriculum itself. And Balaji, the HOD of the Civil Engineering department, had made a presentation just about a few days ago in Mysore, and the Deans and the various HODs, Principal was also there. And basically, he went through what his ideas are on transforming Civil Engineering to a fully AI-integrated curriculum. And he shared some ideas, and I'm going to quickly take you through those slides. Just to give a sort of a template. It doesn't mean that these are the only ideas. Your department, you might have other ideas, and that's completely fine. But since he had put some effort, we thought that, you know, this could be something that we can look at.

**Srikanth**: [13:40] So firstly, he has come up with—I'm going to go through it quickly, Balaji, you can jump in at the end and please add. Semester one to semester eight, everything from the foundational courses that civil engineers need in semester one and two, to the science core that is required for civil engineering in semester three, to the core civil areas, whether it's data analytics or [BIM?] fundamentals using the Revit tool, to semester five on design, semester six advance design, semester seven specialization, and semester eight capstone, which is final year project and so on. So he has actually come up with some ideas on where the AI-integrated new courses and modules fit in, right from semester one to semester eight. I thought this was interesting. So I just thought we'd go through it.

**Srikanth**: [14:34] Secondly, when you come to research areas, he highlighted that there are seven research areas that our NIE Civil Engineering department is focused on, out of which five of them are relevant to AI: structural health monitoring, smart water resource management, intelligent transportation and pavements, sustainable green construction, geotechnical and disaster relief risk AI. Now he's given several topics within those areas. And the two areas are not too much affected by AI, which might be the case in your departments as well. And feel free to say, hey, these areas are where we want to focus on AI with respect to research. These other areas, not now, maybe later, but right now these are the low-hanging fruit, right? Five out of seven in the case of Civil Engineering.

**Srikanth**: [15:25] Final year AI-integrated Civil Engineering projects, there are some ideas. Okay, I'm not going to go through it. Take a look. AI-powered structural audit tool, machine learning-based flood risk analysis using Atlas in Mysore district or greater Mysore, autonomous pavement distress mapper, and so on and so forth. These are all great ideas for AI-integrated projects, okay? In the Apex Committee, I deal with a lot of projects in sustainable cities, agriculture, education, and healthcare. So I'm very familiar with these things, but it's good to come up with some areas where our students can take up these final year projects.

**Srikanth**: [16:07] What tools are needed? Okay, so he's categorized them into structural and design, GIS and remote sensing, so on and so forth, right? And as you go to the right, it gets more generic: generative AI itself, ChatGPT, Gemini, and Claude, and so on. As you go to the left, it gets more specific to Civil Engineering. But the kinds of tools that are needed for us to do this AI-integrated Civil Engineering curriculum, right? Again, very relevant, very useful.

**Srikanth**: [16:37] And one thing that I in fact today morning asked him to add to this: just take one course, Design of Reinforced Concrete Structures, and tell us how it's going to be taught. So here he has come up with a core syllabus, teaching and learning process, research areas, student projects, and key AI tools. And gone into content identification. Remember I spoke about physical plus digital equals "phygital," where **let's use excellent digital content that's already available, integrate that into classroom education, and deliver the best course possible because you don't have to spend a lot of time creating that entire course content.** Somebody already has.

**Srikanth**: [17:24] So he has shown NPTEL and L&T and so on and so forth. Some lab in the UK that has already created these courses, right? And he's also taken this and said, hey, a phygital delivery architecture. Maybe the students can first watch this and then come to class. And then in class, the faculty deepens the understanding, and then there are AI tool exercises perhaps, and so on and so forth, weekly labs. And they can also be, you know, jointly, along with the teacher, maybe we watch too. I don't know, all these we have to try out, right?

**Srikanth**: [18:00] And then for that one course, week one to week 16, he's actually broken up the topics. Physical, digital, virtual labs, AI tools, and assessment. So I'll stop here from a Civil perspective, Civil Engineering perspective. So I thought this was a good first attempt at coming up with something concrete to say that not only do we have clearly what we need to do in eight semesters, but also the research areas that we can focus on. The final year projects that can be taken up which are AI-integrated. The tools and technologies we need, all the way from very specific Civil Engineering tools to generic LLMs. And for one particular course, how do you actually put together the course and deliver the course along with the digital content over the 16 weeks of the course.

**Srikanth**: [18:58] So this is a fairly decent, very good template in my mind. An excellent work. And of course, other department heads are also working on this. I think Tuesday we are going to have the other presentations as well. But I'll stop here for a minute and ask Balaji. Balaji, do you want to share some ideas on what you have built?

**Balaji**: [19:24] Yes, sir. Hope everybody is able to [inaudible]. With respect to this thing, actually this is an overall concept we had a discussion with Srikanth sir and Principal sir. Still we need to refine this thing. Maybe not for entire five modules, maybe few modules where AI integration or some virtual things can be brought in. But again, we need to go back and we need to view this content setup.

**Srikanth**: [19:59] Okay, but excellent work. This is a good template that we can look at, and so on. The other departments, I am looking forward to your ideas, your thoughts on how to navigate this given the larger structure of basic AI and then the department-specific ones. I'm not going to go... the PPT of Balaji is circulated among all faculty, and the HODs now, and Tuesday is the deadline given to the department to present the slides. Wonderful. Wonderful. Look forward to seeing all the other departments and what they come up with as well. Right?

**Srikanth**: [20:39] But this is, I think, it is important for all of us to participate in this exercise. Okay, I want to emphasize this. **Pearls of wisdom don't come from the top. It comes from people who understand the details, who bubble up all the ideas.** And then we can take a look at standardizations, you know, templatization, standardization, and so on. But please get involved. Okay? Work with your HODs, come up with your own department ideas on how to take this further because I mean, we have a wealth of talent here. I mean, I've been looking at all the research papers today morning, I was going through it. They're brilliant, right? So we need to all put our heads together to make this transformation happen.

**Srikanth**: [21:27] Okay, so please take some time. These decks will be available. Work with your HODs. The Deans please coordinate with them, and let's slowly take it forward. It's not important that we standardize prematurely. It's okay to have different kinds of ideas coming from the people who are working with the students, who are delivering, who are working on research and so on. And then we can, as we see wonderful ideas, we can slowly then crystallize and say, okay, maybe we can use these templates and so on. Okay.

**Srikanth**: [22:00] All right. So I'm not going to go through every department right now. We are waiting for them to come back. You know, whether it is ECE, whether it is CSE, ISE, whether it is, you know, EEE, and so on, Mechanical Engineering, and so on. So we will wait. And past Tuesday, we'll go through them and maybe we'll have another session where we can go through them. Okay?

**Srikanth**: [22:25] Now, AI Assisted Research. I don't think I need to go into more detail. Anand has gone into wonderful detail and actually shown practically what is possible. Myself and Professor Nagaraj had worked with the Dean Research Imran, and we had created a spreadsheet that you all contributed to. So we have some data to analyze. But as you can see, there is, we can do this a lot faster. Faster literature review and problem discovery. Like Anand demonstrated: acceleration of data analysis, coding, and experimentation. Like he showed Codex that generated the website analyzing our student feedback. You can allow Claude code or Codex to write code to run your experiments and do the verification much faster. Improved quality through reproducibility and verification. Creation of new domains of research opportunities, which again Anand clearly showcased through his meta-prompting, and how he got that from looking at our Google sheet.

**Srikanth**: [23:37] Okay. Here is one sort of anecdote. The other day, I was at Bata in Mysore on Temple Road, and one of the NIE faculty who was there came up to me and said, "Hey, you know, I was there in one of your lectures," and so on and so forth. And so, you know, I asked him, "What do you do?" and so on. And he talked about what he teaches and the research that he does. And then I gently asked him, "Sir, do you use AI for your research?" And he said, "Absolutely! You know, I use it all the time. And I use it in all these different ways." And he talked about a lot of what I'm putting out there: literature review and, you know, and so on. So I asked him, "So can you compare how long it used to take for you to write a paper without AI, let us say two, three years ago, and now with all these tools that are available, with reasoning, deep thinking, and so on and so forth?"

**Srikanth**: [24:41] **He said it used to take almost six months to write a paper in our context, in the NIE context. And now he says, I think I can finish a paper within one month with all the tools that are available.** Right? That means his effort is pretty much the same, but AI is improving his productivity enormously because of the kinds of things that like Anand has shown us today. Right?

**Srikanth**: [25:07] So I have the same experience in the startups that I am working with, right? The productivity levels are going through the roof with respect to programming and with respect to—not just programming—even on the design side, on the product side, on the QA side, on release management, DevOps, finance, marketing, sales, writing copy for sales and websites and so on. So across the board I'm seeing what this NIE faculty explained that he's seen. Right?

**Srikanth**: [25:49] So we need to go from one paper a year per faculty to at least three or five with the kind of power that we have today. If we don't, others will, and our NIRF ranking will actually fall as opposed to going up, right? So we need to start embracing these things. Okay.

**Srikanth**: [26:08] Moving forward, I'm not going to belabor this point. Now there is one area that comes up. Okay, I was invited by the Ministry of Education to this conference called Akhil Bharatiya Shiksha Samagam. And all the IIT directors and AICTE, lots of the education people were there. From schools, from higher education, engineering colleges, IITs, you name it, all of them were there. Thousands of people literally. And one of the questions, and Professor Kamakoti, who is the director of IIT Madras, he was the moderator. And he asked this question on, you know, how do we use this AI? You know, it seems like... you give a test, you give a quiz, you give a homework, the students are able to ask ChatGPT and get the answer and put it out there. How are they going to learn? Right?

**Srikanth**: [27:11] So there are multiple thoughts around this. So the ethics framework. How do we bring in AI into education but yet not compromise learning? Because at the end of the day, that's what students come for. Okay, that's what colleges do, we teach students, they learn from the colleges. How do you ensure that the learning levels continue to increase even as these AI transformations take place?

**Srikanth**: [27:37] So the one example I gave was calculators. **When I started college, there was a debate whether calculators are good or bad. And there was one predominant opinion that oh, these boys and girls are going to become dullards.** They're going to do all the math using calculators and they're not going to learn how to do math. In those days, we used to do math ourselves in our head, we were so fast, we could do this, that, and so on. Right? So that was a dominant thought. Of course, it's very hard to stop progress. When you get a device that does something, people use it. Same way AI is going to start marching forward.

**Srikanth**: [28:16] So **soon nobody discussed whether calculators are good or bad, because the students migrated from doing arithmetic to algebra, and maybe from algebra to calculus as well.** So they started solving bigger problems. So I think the general opinion was, okay, they're not doing arithmetic, maybe they're not excellent at mental math as people used to be in those days, but they are able to solve bigger problems. I feel that the same paradigm is what we need to apply in education. Now that there are tools like, you know, LLMs that are doing so much more, our students should solve better problems, much harder problems, our research areas should give us enormous benefits in society. Eventually, learning is for that. Right?

**Srikanth**: [29:06] So of course the question it begs is if they don't focus on fundamental concepts and they keep going to AI all the time, will they learn anything at all? And it is a good question. I don't know the answer precisely for that, but what I have seen, I was at MIT for the last Fall 2025 semester—which is August to December 2025—and they're also all grappling with this, right? It's not as though everybody has answers.

**Srikanth**: [29:37] But one thing that's emerging is, you know, colleges are putting out policies as to what is proper use of AI and what is misuse of AI. Academic misconduct. So basically, the general idea that's coming up is do not submit something that is AI-generated as your own. Okay? It's okay to use AI to solve...

---

**Srikanth**: [00:00] To use AI to solve problems and show that hey, actually solved a tough problem. And increasingly, closed book versus open book tests, open book tests are harder because the fact that you can use the book as opposed to testing memory earlier in the closed book. Now, with AI, the problems are going to get even harder. If when we are hiring in our company, we deliberately make the problems so hard that they have to use AI to solve the problem because **we want to ensure that they know how to use AI and they can productively use it and solve the problem quickly**. Right? So similarly, I think we have to come up with an ethics framework around what is good use, proper use, and what is misuse. I will leave it at this. Don't take this as carved in stone. This is just a sample, but as an institute, we have to come up with our ethics framework around AI. Okay. And academic integrity, which applies to all the classes.

**Srikanth**: [01:02] So next is one of the things that I've been pushing within the management committee. The management committee met just a week, 10 days ago at the North campus, and our secretary Uday Shankar took us through all the buildings that are being built right now. Fantastic two buildings, building A and building B, as well as the 14-floor girls hostel, a wonderful campus. But what I was telling them is we need to ensure that they are 100% digital classrooms. Okay? Lecture capture, capture of video platforms integrated with LMS, smart classrooms with audio-video, AV wireless presentation, recording, all of this, taking tests digitally, even if you are sitting in the classroom, coding labs, proctoring, all of this needs to be part of our infrastructure. Okay. You see the kind of power what Anand was able to show when we had digital data. Okay. **The decisions were clear once the data was visualized in ways where the decisions jumped out of the visualizations.** That is only possible if it is digital. If it is all paper, it is going to take a lot of time because we are not going to sit and convert all that paper. The other point I'd like to make is, **with AI, your current work should actually decrease. You can spend a lot more time on improving the quality of learning, ensuring this AI transformation happens**, more research papers, making sure NIE comes right on top as we go through this transformation, because AI can do a lot of the work, as the demos today showed. Okay.

**Srikanth**: [02:56] Phygital. I talked about it. The basic idea is this. For the last 10, 15 years, maybe more, we have seen MOOCs content from edX, which came out of MIT and Harvard. MIT also had its OpenCourseWare, which was amazing stuff. I myself did a bunch of courses in linear algebra and other things using MIT Courseware. And many other platforms. India has NPTEL from the IITs, Coursera has been an excellent platform, and now there is DeepLearning.AI and so on and so forth.

**Srikanth**: [03:31] Now, the only catch is if you ask the question, what is the completion rate on these MOOCs platforms? It is very low. Students don't complete the entire course. **Completion rate is as low as 6.5%.** Okay? Different people give different numbers, but it ranges between 6.5% to 10% is the completion rate. Fantastic courses, the best teachers in the world are teaching courses, but completion rate is low. Why? Because self-motivation amongst youngsters is low. Even with the best content, they are not going to sit alone and understand subjects. It's probably really on the right side of the bell curve, there are going to be probably 10% of the students who are like that, self-motivated and driven, but they will learn anyway, with or without NIE. Right? Now the question is, for the bulk of the students, what do we do? **Can we take this best MOOCs content, integrate it into a physical classroom environment** where there is discussion, there is debate, there is a teacher who answers questions, there is a test, there is some competition, there is that enjoyment of being with a bunch of other students? Right? All of that, if you can integrate it with the best MOOCs content, can we deliver better learning outcomes? And can we make this part of our AI journey because a lot of the AI courses also exist online? Can we integrate this? And that is what you saw, Balaji has tried to do that for civil engineering through his ideas on how this curriculum needs to be delivered in a phygital kind of mode. Okay? So that is the larger idea. And let's all try to figure out how we actually implement this. Okay. Whether do the students have to firstly watch that and come over to the class? In that case, will they actually come prepared? Secondly, maybe we all watch together for 20 minutes, 30 minutes, and then do the discussion around it. Maybe that is still fast, because you didn't have to deliver the entire content. You could let the video do it, and then you could allow people to ask questions and have a discussion in class and so on. So what is the best way to do it? I think you people will find out what the best model can be. Okay. And there are some lots of courses across various, whether it's civil and so on.

**Srikanth**: [06:07] So theory versus practice. I'll leave, this is the last idea, and then we'll close. So... Hello? Was there a question?

**Unsure**: [06:17] Probably not.

**Srikanth**: [06:19] Alright. So theory versus practice. I came and spoke about my experience at MIT taking one of the classes called "How to Make Almost Anything". And, of course, it was many years since I had gone back to college and taken a course and so on, but this was an incredibly hard course, and we had to do a lot of things. Every week we had to actually build a system. In fact, what I built as my final project is right here. I built this AI assistant that will help me sort through all my emails, my calendar appointments, the weather, and tell me what to do next because I get a lot of messages every minute. And I built this, including the... can I show you... okay, including the Raspberry Pi 2040 controller and so on. But long and short of this was the fact that **I learned 10x more than what I would if I did only theory.** Even the Raspberry Pi 2040 that I used, if I had just read the data book—by the way, I am a chip designer, I have done design of Intel CPUs and chipsets and so on, I know this subject quite well because it's been a long time since I did it—but if I had just read the data book of RP2040, which was the microcontroller that I used, it would have been nothing. I wouldn't have learned that much. When I had to design the PCB and mill the PCB and solder it and get all the peripheral chips to work—because I needed a mic, I needed a speaker, I needed to drive a matrix display—when I had to integrate and actually make it all work, build the code for it in Python, a language I didn't know, and soon I learned how to use IDEs and so on, **I learned at 10x, 50x the pace that I usually learn.** Okay? In those three months, I learned more than many years of my work. Okay? So what it tells me is that if students along with faculty do projects—okay, it doesn't always have to be physical projects, it could also be AI projects, it could be coding exercises, it could be various things in civil and mechanical and so on—but when they do things, they learn a lot. And that to me is more useful in the industry than theoretical knowledge. Theoretical knowledge is cheap now. You ask ChatGPT, it'll give it to you in a fraction of a cent. Anand showed intelligence is very cheap. Information is very cheap. Okay? **Memorizing that information is not going to get you jobs.** Okay? So in that era, building, looking at larger systemic thinking, how do you put this device together? How do you build the code? How do you get the chip to work? What are the peripherals? How do you do the PCB? How do you do the display? Can it connect on the internet? Can I call NTP servers? Can I call other tools? All of those kinds of ideas are what will get our students prepared for the industry. Right? For that, just memorization and the textbook is not enough. We have to teach them how to build systems. So they need to experiment and start getting a deeper understanding of things, not just passively reading texts. So conceptual understanding, design, troubleshooting, iterating through this, these are more powerful ideas.

**Srikanth**: [09:53] So what I suggest is course projects. Even value-added courses, small small projects along the way. Of course, the final year capstone projects, super important. We've always done it even from my time there. Internships are principal. Dr. Nagendra has written a book on internships. He's done a lot of work at BITS Pilani on how to make these internships really work, not only for our students but also for the industry. Can't be a one-way street. Industry is not going to keep doing favors. Our faculty need to get involved. It's not enough to say, "Ah just go, I'll give you students, you do whatever you want with them." That does not work. I keep hearing this from the industry over and over again. Industry projects. You want to understand practically what is needed instead of sort of theoretical ideas, industry will tell you when you work with them. And of course, startups, the entrepreneurial ecosystem. All of these are ideas towards doing, not just reading, not just theory, right? So I think this transformation, this is the third part, is also super important.

**Srikanth**: [10:53] That's pretty much it. So the next steps, our principal is taking this forward. I mean, these are some ideas that we put down, but since then, can I ask Dr. Nagendra to tell us a little bit about the next steps?

**Dr. Nagendra**: [11:13] Sir, the next step is, you have to be presented in a like Balaji has made a presentation. With Dean Academic and the input what we got, we have a plan how we can take it forward now based on the presentation and what we had last week. And we are meeting once again on Tuesday. Monday or Tuesday we are meeting with all HODs where they are going to have their presentations. Post that, now I actually given you a tentative plan on that day that HODs prepare presentation on that day around 14th, that how actually we can review that strategic or action plan, and we will come back to you, that how we will take it forward. And, now they are, frankly speaking, all the faculties are working with identifying the areas that can be integrated into the curriculum or to start projects, mini projects, things like this where they can use. We are not going for structural change, we are going for course structure change. Where say for example, in the practice component instead of doing theory courses, this can be converted into practice spaces where we can convert a theory course into a practice-based component and integrate AI tools into it. That's one area they are working. And how to integrate curriculum in mathematics, we recently shared you one table also on articles. That how actually we can run down the base of mathematics, how actually we can work on that. That is another thing what maths HOD and Dean Academic are working now. So these are all the steps that it's been initiated very quickly and strategic plan I'll revert back to you with that. And Mr. Anand... okay, that's the next steps sir.

**Srikanth**: [13:00] Okay. Wonderful, wonderful. Thank you so much Dr. Nagendra. I was also going to say, by the way, the management committee, I am pushing hard for this AI transformation. In fact, the management committee requested me to drive this effort and come up with ideas and do a presentation on how NIE has to look at the AI transformation, and that's how this whole thing got started. In fact, the secretary Uday Shankar requested me to...

**Unsure**: [13:32] Sir, your microphone is muted.

**Srikanth**: [13:35] Oh sorry. Okay, yeah, I was having some echo here, so I turned it off. Okay, so basically what I was saying was, the management committee approached me, I am in the management committee. In fact, our secretary Uday Shankar approached me and said, can you give a talk on the transformation that NIE needs to take up for this AI revolution that is taking place. Okay. And that's how it started. And I did a presentation along with Anand about I don't know 10 days ago, it was at NIE at the North campus, along with HODs and faculty, and Mr. Narayana Murthy had also joined us. And that was a start. But what I will also tell you is, we had a board meeting, again, on the same day, earlier. And I talked about what we have to do and also talked about, we need to fund this. Okay. We are talking about infrastructure, we're talking about an entirely digital campus, Wi-Fi everywhere, we are talking about various tools and so on. And by the way, Balaji's deck had a request for about 3 to 3.5 crores for civil engineering department itself. Right? And, of course, while we will definitely look at all of this because we need to take this up very seriously. The management committee is very serious about this and I am going to represent this in the management committee. We have to be careful, of course, and make sure that whatever we buy is useful across the board. It shouldn't end up like servers that we have bought that, it's hardly been used, sit there. So it has to be judicious. But I would say, feel free in your decks to suggest the really important tools and technologies that you need for this transformation. We will certainly look at it. We will certainly try and see what we need to buy and what we need to install and how we need to change the infrastructure in order to get through with this AI transformation. I'll stop here. Any questions you have, to me or to Anand, please feel free, we can go through some Q&A.

**Dr. Nagendra**: [16:07] Mr. Anand.

**Anand**: [16:09] Yes, please.

**Dr. Nagendra**: [16:10] Can you please share the feedback survey of students and research link whatever you have presented?

**Anand**: [16:19] Absolutely.

**Dr. Nagendra**: [16:21] Because my work starts from there.

**Anand**: [16:26] Definitely.

**Srikanth**: [16:38] Anand, could you host what you have created, that website internally at NIE and then keep improving on that?

**Anand**: [16:54] Exactly.

**Srikanth**: [16:56] Any other questions? Comments? Feedback? Thoughts? From the other HODs, how are your plans coming along? Maybe mechanical HOD Prakash, are you there? Or what are your plans coming along on this AI transformation? Other HODs, anybody who want to share your thoughts, any doubts, any questions as you put your plans together?

**Unsure**: [17:42] Yes sir. From EC department, we have divided our faculties into four different streams like VLSI, embedded systems, communication, and signal processing. And based on that we are working on how to integrate AI into these different core streams. So we will come out with a plan on Tuesday.

**Srikanth**: [18:07] Wonderful, wonderful. Look forward to the plan and there's quite a bit that you can do. I am from that department. And like I was giving an example of my MIT experience, you can do quite a bit with AI. So I look forward to seeing your plan on Tuesday.

**Unsure**: [18:27] Sir, from Electrical and Electronics, yes sir, we are working on that. So we have identified some subjects where AI concepts can be added. Right from the basic AI literacy, as you are frankly speaking since 15 days, and even the research areas basically where it can come out. So those things we are working among the faculties and we will present it on Tuesday sir.

**Srikanth**: [18:55] Wonderful. Look forward to your deck.

**Dr. Nagendra**: [19:01] Mr. Srikanth and Mr. Anand, one point and challenge on day one is our faculty getting aware of the integration of AI tools what all we presented so far. So the major challenge for us is to go for a faculty development program where we create their mindset on doing this. Now they can come out with a practical applications on projects like this, but how deep we can integrate into the curriculum unless we train them a lot. We look out to see how we can fix this down the line in the next 20 days or one month.

**Srikanth**: [19:39] Sir, I ask you, the course that you offer, and even the parts of what you did was quite amazing. I mean I learned a lot myself. Is that something that the faculty at NIE can join? The classes that you offer? And use the tools...?

**Anand**: [19:59] I'd started building another course for IIT Madras faculty specifically. It's literally AI for faculty, both for their use as well as for use in education. I'll be releasing a version of that in about two weeks. So yes, would be very happy to run that in parallel at NIE as well.

**Srikanth**: [20:25] Oh, that will be very useful. Wonderful. And this we will coordinate with you on making sure.

**Anand**: [20:34] Sure.

**Srikanth**: [20:36] Any other questions?

**Anand**: [20:44] There is a question Srikanth on the chat. Zoom chat.

**Dr. Nagendra**: [20:50] Dr. Divakar is messaging something. Dr. Divakar messaged something that in the AI department, mechanical department, they did twice brainstorming session on this. He messaged it.

**Srikanth**: [21:07] He says twice we did brainstorming session in the department, we are finalizing shortly. Wonderful, wonderful. We look forward to your ideas and presentation Dr. Divakar. Okay.

**Dr. Nagendra**: [21:27] Great presentation sir, especially the analysis what you made, the way you taught us how to use the AI tools for research and everything is quite amazing. And thank you, Mr. Srikanth. You are always supportive for us. In this type of management, you are standing behind the principal all the time at every step. I think we can do wonders. And this is a really bright, heart-touching presentation that what both of you have made. And on behalf of my faculty, I thank both of you very much.

**Srikanth**: [22:03] Thank you. Thank you Dr. Nagendra. And Anand, thank you so much for being here and I was quite sure that you're gonna amaze all of them with the analysis that has been sitting under our nose all these months and you just showed what AI can do with this data and how we can actually make better decisions and improve NIE. Thank you so much for joining today.

**Anand**: [22:30] Pleasure.