↑ Click to open full-size sketchnote in overlay · Also opens in new tab
NIE Mysore · 21 March 2026 · Faculty Session
How Srikanth Nadhamuni and Anand turned NIE's own hidden data into a live consulting engagement — and what 2,500 students had been silently trying to say for years.
Read the full transcriptThe Setup
It was supposed to be a simple Saturday morning webinar. Srikanth Nadhamuni — NIE alumnus from 1984, the man who built the Aadhaar technology, co-founder of Trustt — was presenting to the faculty of his alma mater from Bangalore. Anand was dialling in from Singapore. The topic: how AI would reshape engineering education. The mood: anticipatory, slightly formal.
Then, about eight minutes in, someone typed in the chat: "Faculty are getting the message that the meeting is at the host's allowed capacity."
There were more than a hundred faculty members trying to join, and Srikanth's Zoom licence — the business kind, the expensive kind — had a cap. In full view of the audience he was about to inspire, Srikanth did the one thing that proved his thesis better than any slide: he opened ChatGPT and asked it what to do.
"Let me quickly ask ChatGPT what to do. Give me one second."
— Srikanth Nadhamuni, approximately eight minutes into a talk about AI
The solution was to spin up a parallel Google Meet session, which Anand hosted, while Srikanth streamed audio on both. For the next ten minutes, two veterans of India's technology establishment fumbled with mic echoes, muted tabs, and a Google Meet that only some people could hear. It was gloriously human, and in its own way, deeply instructive. The old infrastructure — technical and institutional alike — was straining under new demand. That tension would be the subtext of everything that followed.
Act I
Srikanth began where all honest AI stories must: with the long decades when almost nothing worked. The term "artificial intelligence" was coined at a conference at Dartmouth in 1956. And yet, as recently as 2022, most people had never had a real conversation with a machine. The field spent the next six decades promising a future it consistently failed to deliver.
"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. How is it even possible? Because for millions of years nobody could do this excepting human beings."
— Srikanth Nadhamuni
Srikanth had a personal story here. Decades ago, flying to the United States to do his master's degree, a stranger on the plane asked what he planned to study. "I want to do artificial intelligence," the young Srikanth said. The stranger nodded. The years passed. AI went through its winters. Srikanth built chips at Intel, built Aadhaar for a billion people, built companies in Bangalore. And then, in his sixties, the thing he had dreamed about on that plane finally arrived — not at the edges of possibility, but at the centre of everyday life.
The Slides
Selected slides from the presentation Srikanth shared with NIE faculty. Click any image to expand.
Act II · LLM Pricing
When Anand took over the screen, he started with a chart. Each dot was a language model. The x-axis was cost. The y-axis was intelligence.
In 2023, the best model available — the model that could do things that genuinely astounded people — cost roughly $8 per million tokens. One million tokens is about the length of the entire Harry Potter series. For $8, you could feed the whole saga to the model and ask it anything.
"Instead of one college graduate, for the same budget, I can hire 60 college graduates."
— Anand, on what happened to LLM costs between late 2023 and mid-2024
By May 2025, a model called Gemma 3 could do the same — smarter, actually — for two cents. The same task that cost $30 for GPT-4 in March 2024 cost two cents fourteen months later. Not cheaper. Not ten times cheaper. A thousand times cheaper.
The intelligence, meanwhile, had climbed from roughly "bright high schooler" to "tenured professor." For $20 a month, you had a full professorial mind available at 3 a.m. on a Saturday, with infinite patience, zero ego, and a read through every paper published in your field.
Act III · Live Analysis
Here is where the talk became something else. Something more uncomfortable, in the best possible way.
Anand had a thought experiment. What if you hired McKinsey — the global consulting firm — to do a strategic engagement for NIE? What would the brief be? Simple: How can we improve our NIRF ranking? What factors matter? What does NIE's specific data say?
That engagement would cost, conservatively, several hundred thousand dollars, take three months, and produce a deck with lots of muted-blue slides. Instead, Anand fed the question into Claude Sonnet 4.6 with extended thinking enabled — a model at roughly PhD-candidate capability — and let it run.
"You effectively have the equivalent of a tenured professor sitting in your pocket for $20. Worth exploring."
— Anand
The result was a structured breakdown of NIRF's five pillars, their weightages, and NIE's relative performance on each. The answer was clear, and it was not flattering. NIE had started from around Rank 100 in 2018 and slid steadily to around Rank 200 by 2022 — before a partial recovery to around 151. The biggest gaps: Research and Professional Practice, and Peer Perception. The biggest strength: Graduate Outcomes.
But the analysis did not stop at diagnosis. Anand had also fed in NIE's internal publication tracker — a spreadsheet that captured which faculty, from which department, had papers in which stage of development. The AI combed through it and began generating surgical, department-level recommendations: which citation clusters to build, which research gaps to fill, which grant agencies to pitch, and which corporate CSR-to-lab routes were most likely to convert.
"Identify the underpublished niches where you can start building your citation clusters directly."
— Anand, quoting his AI analysis back to the faculty
A specific example: Ramya, a researcher in the ISE department, had produced 18 papers — the highest individual output in the dataset. Several colleagues had significant outputs too, but they were working in parallel, not together. The AI's suggestion: build a coordinated citation cluster. Publish papers that cite each other's work, not because it is gaming a metric, but because coordinated research produces stronger, more referenced work. This is what every successful research department at top universities does. NIE just had not visualised it yet.
And then Anand said something that made people pause:
"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."
— Anand
Half an hour. Yoga. And when he returned from his mat, an institution's strategic landscape had been mapped, annotated, and organised into actionable priorities. The consultants at McKinsey have been replaced not by a robot, but by a process: feed good data in, ask the right questions, get expert-level output out. The human part — knowing which questions matter, knowing which outputs to trust — remains. But the heavy lifting has moved.
The Method
Srikanth asked Anand a pointed question: how did he know what to ask? Did you need Anand's particular expertise in data visualisation to replicate this? Or was it cut-and-pastable?
The answer was both liberating and slightly disorienting. Anand did not write the data analysis prompts himself. He first asked Claude: given the data I have, what are the most valuable analyses I could run that would showcase AI's power and improve NIRF rankings? Give me a dozen options, rank them, and give me the top three as ready-to-use prompts.
Claude did the thinking. Anand copied the prompts and ran them.
"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... If you know, prompt. If you don't know, meta-prompt."
— Anand, on his approach to AI-assisted work
The parts where Anand did bring his own expertise were the visualisation and storytelling instructions. His data-story methodology — condensed into what he calls a "skill file" — specifies things like: write like Malcolm Gladwell, visualise like the New York Times graphics team, use tooltips for context, use pop-ups for citations. That skill file, he noted, is cut-and-pastable by anyone, regardless of whether they know anything about data visualisation.
"My expertise, so to speak, is condensed into one data story skill," he said. "I look at the output and say, no, change this, change this. But there are so many areas where I have no clue. I know nothing about NIRF. There, I will just take its expertise — and it is anyway at a tenured PhD level and above."
Act IV · The Hidden Signal
There was a second dataset. NIE had been collecting student teaching feedback for years — 2,500 responses across multiple cohorts, rating faculty preparation, lab quality, course structure, accessibility, and more. Most of it had been processed into numbers and averages. The open-text comments — "students wrote about broken lab computers, they wrote about instructors who were reading from the slides, they wrote about the 85% attendance rule" — had largely sat unread.
Anand fed the entire dataset to a coding agent. He wrote a detailed prompt (itself generated by Claude) instructing the agent to merge the files, normalize the columns, compute per-department per-cohort statistics, run a cohort comparison, and surface the themes from the qualitative text.
"The teacher in the room is everything."
— The first and most important finding from 2,500 student feedback responses
The headline finding was stark. Of all the factors students rated — lab quality, course structure, lecture pace, faculty accessibility — only one had a statistically strong correlation with overall learning satisfaction. Not lab equipment. Not course design. Faculty preparation. A 71% correlation. Everything else was secondary.
The Data
The gap between first-year enthusiasm and senior disillusionment mapped neatly onto lab quality: outdated computers, MATLAB crashing, mechanical labs with broken equipment. Students had been writing this in their feedback forms for years. Nobody had aggregated it with enough statistical power to see the pattern clearly. A coding agent, given the right prompt, did it in thirty minutes while its operator held a yoga pose.
The qualitative feedback was equally pointed. Students were asking for less slide-reading. They wanted board work, interactive engagement, more real-world examples. One cluster of responses was particularly striking:
"Board teaching is more effective, please use it."
— A recurring theme in the student feedback text, as extracted by the AI
And then there was the attendance rule. The 85% attendance mandate, which exists to ensure students show up, was in multiple responses being cited as the reason students had missed placement interviews. One absence too many, and you lose exam eligibility. No eligibility, no placement. A rule designed to protect learning was, for a meaningful subset of students, actively destroying their career prospects. It was the kind of feedback that gets buried in averages. The AI surfaced it to the top.
Anand was careful to note that not every complaint reflects an objective truth. But when a theme appears repeatedly across 2,500 responses, across multiple cohorts, across multiple departments — it is no longer an outlier complaint. It is an institutional pattern.
An Interlude
Mid-presentation, Srikanth told a story. He had been shopping at a Bata store on Temple Road in Mysore — the kind of quiet, everyday errand that yields no particular insight — when a faculty member recognised him and introduced himself. They got to talking. Srikanth asked, as he does: Do you use AI in your research?
The faculty member's answer was enthusiastic. He used AI for literature reviews, for identifying research gaps, for drafting. Not experimentally. Routinely.
Srikanth pressed: before all these tools, how long did it take you to write a paper?
"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."
— Srikanth Nadhamuni, recounting the Bata conversation
The same human. The same brain. The same institutional constraints. One-sixth the time. The research productivity implication is staggering: if every NIE faculty member halved their paper-writing time, the institution's research output — which directly feeds NIRF's 30%-weighted Research and Professional Practice score — could double or triple without adding a single hire.
Act V · The Roadmap
When Srikanth returned to his slides, the theoretical stage was set. Now came the prescription. What should NIE actually do? Three ideas, in order of precedence.
The phygital idea addresses a specific and under-discussed failure. Over the past fifteen years, tens of millions of dollars have gone into producing some of the finest instructional material the world has ever seen: MIT OpenCourseWare, Harvard's edX courses, India's own NPTEL, Coursera, DeepLearning.AI. The best professors on the planet, recorded in HD, freely available to anyone with an internet connection.
And the completion rate hovers around 6.5%.
"Fantastic courses, the best teachers in the world are teaching courses, but completion rate is low. Why? Because self-motivation amongst youngsters is low."
— Srikanth Nadhamuni
The MOOC revolution's great unfinished business is motivation. Digital content scales infinitely; human attention does not. But put that same content inside a classroom — where there is a teacher to field questions, peers to compete with, tests to prepare for, and the social friction of attendance — and completion rates climb dramatically. Srikanth's insight was simply to stop treating physical and digital as alternatives and start treating them as complements.
The Civil Engineering HOD Balaji had already done the practical work: a semester-by-semester AI-integrated curriculum for his department, complete with NPTEL course links, week-by-week delivery plans, final year project ideas (AI-powered structural audit tool; machine learning flood risk mapping of the Mysore district), and a ₹3–3.5 crore budget ask for the necessary infrastructure. It was the most concrete piece of institutional planning Srikanth had seen, and he held it up as the template every department should follow.
The final idea — the shift from theory to practice — was illustrated with a personal story. Srikanth had spent a semester at MIT in late 2025, taking a course called How to Make Almost Anything. Every week, the students had to build something physical — not design it, not write about it, but actually build it. His final project was an AI assistant to help manage his overwhelming inbox and calendar: a custom PCB with a microcontroller, a microphone, a speaker, and a matrix display, all connected and programmed in Python, a language he barely knew.
"I learned 10x more than what I would if I did only theory. 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 — when they do things, they learn a lot."
— Srikanth Nadhamuni
The argument is uncomfortable for institutions built on lecture-hall pedagogy, but the evidence is hard to ignore. Theoretical knowledge, Srikanth observed, is now practically free: "Ask ChatGPT, it'll give it to you in a fraction of a cent." What is expensive — and therefore valuable — is the ability to take that knowledge and build something real. Systems thinking. Debugging under pressure. The willingness to fail, iterate, and try again.
"Memorizing that information is not going to get you jobs."
— Srikanth Nadhamuni
The Oldest Argument
One question hovered over the entire session, unspoken but present: if students can use AI to answer any question, will they learn anything at all? Srikanth had been asked this at the Ministry of Education's Akhil Bharatiya Shiksha Samagam, with IIT directors and AICTE in the room. He had a ready answer, and it was historical.
"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."
— Srikanth Nadhamuni
That debate lasted about five years. Then it quietly ended, not because people decided calculators were harmless, but because something more interesting happened: students stopped doing arithmetic and started doing algebra. And from algebra to calculus. The floor of what was worth learning rose, and with it the ceiling of what was achievable.
The same is happening now, faster. AI does not just calculate — it writes, codes, analyses, and reasons. So where does the floor move to? Srikanth was honest: "I don't know the answer precisely for that." But from what he had observed at MIT — where the same question was being debated by people who had no more certainty than anyone else — the direction was clear: harder problems, not easier ones. Open-book exams that require AI to solve, because the problems are hard enough that AI is necessary. Assessment of judgment, not memory.
In his own companies, hiring processes had already adapted: "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." The credential being evaluated is no longer knowledge. It is the ability to direct intelligence — artificial and human — toward problems that matter.
The Detailed Roadmap
The full presentation covered curriculum design for every department, AI ethics frameworks, phygital delivery architecture, and research acceleration. Click any slide to expand.
Key Takeaways
Further Reading
The live analyses Anand ran during the talk are publicly available. These are not summaries — they are the actual AI outputs that were shared on screen with NIE faculty.