In a small village town surrounded by jungles in West Bengal, a young boy grew up knowing Newton’s laws before most children learned to read chapter books. His father, a physics teacher, filled their home with scientific curiosity – even if the answers sometimes stopped where the books ended. Today, Subham Ghosh is tackling one of materials science’s most pressing challenges: making decades of research accessible in hours instead of months.
The Homesickness That Changed Everything
Subham’s journey took an unexpected turn early. After securing admission to prestigious Jadavpur University for engineering, homesickness pulled him back. “In the village, you see the same people every day since childhood. Suddenly too many people—I couldn’t handle it.”
He returned home for a BSc in Physics, following his father’s footsteps. But something had shifted. When he qualified for IIT through JAM, he got into IIT Kharagpur – close to home. He chose IIT Mandi in Himachal Pradesh instead. “If I go to Kharagpur, I’ll come back home again. So I chose Himachal – there’s only one way to go, I can’t come back alone.”
Two years in the mountains transformed him. He found friends from across India, worked as a JEE coach, launched a startup in Jaipur focused on 3D advertising (still running today), and discovered his calling at the intersection of AI and physics.
The Problem Worth Solving
Imagine developing new eyeglass materials – flexible, durable, bend-resistant. To understand what’s been done before, you would read hundreds of research papers, extracting compositions, properties, processing parameters, structural characteristics. It could take months.
This is what Subham is solving.
As an IndiaAI PhD Fellow at IIT Roorkee’s Mehta Family School of Data Science and Artificial Intelligence, his research in materials informatics applies AI to the materials domain, but not simply. “We can’t apply ChatGPT normally to materials,” he explains. “Molecules have chemical formulas with Na, proportions, isotopes. The AI needs to understand scientific strings, not everyday language.”
He is building comprehensive datasets from materials science literature- not just collecting data, but creating intelligent systems that understand scientific context. His recently published paper in Computational Materials Science (a Q1 journal, top 25% of publications) introduced a thermoelectric materials dataset with over 11,000 entries, created in just ten days.
But that is just the beginning.
The Billion-Data Vision
“We’re building an AI agent workflow that will extract datasets from anything in materials science,” Subham explains.
The vision is audacious: systems where researchers input hundreds of papers and receive comprehensive, structured datasets in hours. Where AI agents don’t just collect information but understand physical laws, automatically apply boundary conditions, respect scientific constraints.
“Existing genAI knows physics laws, but we have to prompt it – restrict this, restrict that,” he notes. “I want to build something that restricts things automatically based on physical laws.”
The challenge? Scale. “If I go through all of materials science, it will be billions- more than billions. I don’t have that many API keys.”
But the opportunity is transformative. Outside India, in the UK, France, and US, startups are securing significant funding for AI in materials science. Subham is positioning himself at this frontier, building foundational datasets and models that make innovation possible.
His plan? “Hopefully within one and a half years, I’ll deliver easily.” Then? “Research-oriented startup. Things are changing so fast and we can’t stick to one particular thing. Research-oriented, modern startup – that’s the best option.”
Learning in the Age of AI
Subham’s approach reflects how AI is changing knowledge work. “AI is very broad – everyone works on narrow lines. Everyone doesn’t know everything. That’s why I use ChatGPT constantly. When new papers come out, I upload them and ask, ‘What technology are they using?'”
He’s part of a small but growing AI community at IIT Roorkee, working with Professor Abhishek Tewari, a mentor with an impressive background who offers the perfect combination of rigor and freedom.
“Sir is chill about things,” Subham notes. “He lets you explore and doesn’t say you have to do one particular thing. We have presentations every 15 days, ensuring our presentation skills improve.”
The environment suits his creative process. “No one tells me to come to lab at 10 AM. I wake up at 1 PM if I want. I’ll go at 3 PM and work through the night. I work more than corporate guys, but there’s no deadline. I can do things whenever I want.”
The Challenges of Pioneering
Publishing groundbreaking work isn’t easy. “The most difficult part is convincing people what I’m doing is novel and good for that journal,” Subham admits. “Publishing papers is extremely essential to securing my PhD.”
The process requires multiple convincing stages: your supervisor, then demonstrating results, then journal standards, then surviving peer review from worldwide experts who may question everything.
“The relationship with your mentor has to become very strong,” he emphasizes. “You have to be very connected with your supervisor.”
Yet he remains committed. “For the sake of knowledge, we can do PhD. And if you’re innovating something very far from the world, you can even win the Nobel Prize. That’s my childhood dream.”
From Homesickness to Home Runs
The journey from a village where a boy struggled with homesickness to IIT Roorkee where a researcher builds AI systems that could transform materials science is remarkable. But perhaps more remarkable is how Shubham has maintained his authenticity throughout – the humility about his presentation skills, the honesty about his journey, the humor about working nights and drinking chai three or four times daily with labmates.
His father’s physics books may have had their limits, but Subham now works at the frontier where new knowledge gets created, where AI agents will soon extract insights from billions of data points, where materials scientists worldwide might access in hours what once took months.
Subham is a PhD researcher at IIT Roorkee’s Mehta Family School of Data Science and Artificial Intelligence, working at the intersection of AI and materials science. His work on materials informatics aims to revolutionize how researchers access and utilize materials science knowledge.
Published paper: https://www.sciencedirect.com/science/article/abs/pii/S0927025626000406