Kathmandu, Nepal · Independent Researcher
Language is not just communication —
it is the architecture of thought.
I build systems that understand the world through the words
that most of the world still speaks in silence.
I grew up in Nepal noticing a quiet gap between the complexity of our problems and the simplicity of our tools. Nepali heritage sites drowning in unread visitor feedback. Farmers turned away from banks because their lives couldn't be translated into a credit score. Languages spoken by millions, yet invisible to the machines reshaping the world.
I didn't set out to become a researcher. I set out to fix things. Somewhere between fine-tuning a small language model and watching a multi-agent system route a farmer's loan application for the first time, I realized this is research. Quiet, stubborn, necessary.
My curiosity lives at the intersection of language, machine intelligence, and the people those two things tend to overlook. I believe the most important AI research is not the loudest. It is the most locally rooted.
Most of the world doesn't speak English. Yet most AI does. I work on making language models that understand how real people especially Nepali speakers actually write and think, including the Romanized scripts they use every day.
Intelligence doesn't require scale. I explore small language models that can run in resource-limited environments because meaningful AI shouldn't require a data center to be useful to people who need it most.
Complex human problems like credit access or heritage preservation need systems that reason across multiple steps. I build those systems for contexts that actually matter to communities close to home.
Technology is only as good as the people who understand it. I care deeply about helping communities engage with AI thoughtfully to use it as a tool for thinking, not to surrender to it as an authority.
“The most dangerous illiteracy is not being unable to read — it is being unable to question what you read.”
I research because I've seen what happens when powerful tools arrive in communities that aren't prepared to question them. AI is already reshaping how loans are granted, how knowledge is surfaced, how voices are heard. If the people closest to these problems don't build the solutions, someone far away will and miss everything that matters.
Nepal is not a problem waiting to be solved by foreign technology. It is a living context rich in language, complexity, and human intelligence waiting for systems that actually understand it. That belief is why I research. That belief is what I research.
Nepali speakers communicate every day in Romanized script k cha?, thik cha yet no accurate model exists for this lived language. I'm fine-tuning a small, efficient model trained on how people actually write: a bridge between living language and machine understanding, built at the edge of what resource-limited AI can do.
Millions of Nepali farmers can't access loans because traditional credit systems don't understand their lives. This five-agent AI system reads documents, evaluates income patterns, and makes fair, explainable credit decisions in a country where a bank branch might be a day's walk away. Selected at the Global IME Hackathon for its potential to democratize financial access.
An education reformer who asked what school systems were actually producing not grades, but human beings. His insistence that institutions serve people, not compliance metrics, is the same lens I bring to AI systems. What kind of person does this technology produce? What does it ask of them?
The questions that pull me hardest: what does it mean for a machine to understand? How do words carry culture across generations? Philosophy keeps research honest it demands that we ask why before we ask how.
Technology is political. Who controls the models shapes who has voice, access, and power. Reading history and geopolitics reminds me that research is never neutral every dataset encodes a worldview, every model reflects a choice about what matters.
A future where a farmer in Sindhupalchok can access a fair loan through a system that speaks his language literally and culturally. Where a student in Kathmandu uses AI as a thinking partner, not a shortcut. Where the models that shape the world were shaped, in part, by the people who live in it.
Not utopia. Just equity. Just honesty. Just tools that work for everyone even the millions whose words the machines have yet to learn to hear.
If you're working on low-resource language AI, socially grounded research, or AI literacy I'd genuinely love to hear from you.