Research
How do we make machines understand language?
NLP · Telugu · graphsone sec — coffee first
Product Manager · Intuit
I build products at the intersection of technology, users, and AI.
Product Manager at Intuit with a background in software engineering and NLP research. I enjoy turning ambiguous problems into products that are useful, technically grounded, and worth building.
scroll downI studied computer science at IIIT Hyderabad, built software at Goldman Sachs, and now I work in product at Intuit. The questions I care about are why people are struggling, what is actually going on under the request, and what we should not build.
I'm a builder at heart. I'll happily talk to just about anyone, and half my best ideas come out of a random conversation. Turns out "let's just try it" is basically my whole personality.
I started with code. It changed how I think about products. Engineering taught me systems and trade-offs. Product taught me to start with the person living the problem.
Not a career ladder — stamps in a passport. Same traveler, different cities.
How do we make machines understand language?
NLP · Telugu · graphsHow do we build reliable systems around messy problems?
code · constraintsWhat problem is actually worth solving?
users · judgmentWhat becomes possible when machines can reason, understand, and act?
still collecting stampsThat's the space I'm interested in now.
Taking vague customer pain and turning it into a problem that's actually defined.
what's really going on?Understanding the messy workflow, then making it simpler for the person using it.
less spaghetti, pleaseWhere AI, agents, and intelligent workflows can actually change how the product works — not just sprinkle a chatbot on top.
if it needs a wrapper, skip itGoing from a blank page to something tangible enough to test, learn from, and improve.
ship a sliceprofessional
Product at Intuit. Engineering at Goldman Sachs. Research at IIIT Hyderabad. Same curiosity, different questions.
QuickBooks / Intuit Enterprise Suite — Projects & Extensibility · Bengaluru
Commercial Real Estate Mortgage Platform · Bengaluru
Owned the enterprise-readiness roadmap across 5 workstreams. Used 5K+ customer signals to name reporting as the #1 pain, then drove 30% of identified opportunities to scope or launch.
Shipped recommendations trained on historical transaction patterns — 33% acceptance across 325+ mid-market companies, automating a third of dimension assignments.
Co-launched a no-code data modeling platform so 4.8K mid-market customers can extend business objects across core workflows without engineering.
Led 0-to-1 for a mid-market agent that lets customers set up projects, build budgets, and track costs in natural language. Paired that with discovery on construction job-costing and payroll.
Built an internal asset-management product with a maker-checker workflow (−32% error-correction time) and a third-party integration PoC (−40% manual reconciliation).
B.Tech + MS by Research in Computer Science · Gold Medalist for All-Round Excellence
My interest in AI didn't start with ChatGPT. It started in a lab at IIIT Hyderabad's Language Technology Research Center, trying to get language technology to work for Indian languages that most models barely notice — especially Telugu. Datasets, embeddings, classification, graph networks, summarization… the unglamorous stuff that actually makes language systems exist.
That work still shows up in how I build products: start with the constraint, care about the data, and don't confuse a good benchmark with a good experience.
What happens when you try to build NLP for languages that don't get English-sized data?
Datasets, embeddings, graph-based models, and multi-task text classification.
Unsupervised summaries for Indian languages — useful output when labeled data is scarce.
Same curiosity about language, models, systems, and people — now aimed at products people actually use.
Unsupervised graph-based summarization for Indian languages
How do you automatically write useful summaries when you don't have English-sized training data? We used graphs to pull signal out of the language itself.
read paper →Graph convolutional networks for large-scale low-resource languages
Can graph-based approaches help classification across several NLP tasks when the language is resource-poor?
read paper →Am I a resource-poor language?
Datasets, embeddings, models, and a look at what Telugu can (and can't) do across a handful of NLP tasks.
read paper →Full list lives on ResearchGate — the story lives here.
thinking · building · the rest of me
Notes I keep coming back to. Not a blog — more like the sentences I scribble in the margin of a spec.
People ask for a button. Underneath is usually a broken workflow, a missing piece of context, or a job they can't get done.
Not the model. Whether it changes a real step in someone's day — and whether you can tell when it's wrong.
When the work is multi-step, messy, and the user shouldn't have to babysit every click. Not because agents are trendy.
A prototype proves it can work. A product survives the boring parts: edge cases, trust, evals, and "what if this is wrong?"
I used to start with the system. Now I start with the person, then go back to the system with better questions.
Roadmaps full of answers are easy. The work is naming the question that's actually worth the next quarter.
I learn products best by actually building them — even the tiny, slightly chaotic ones.
AI-native products that solve a real job, not an LLM wrapped around an existing workflow.
status: poking aroundBought a domain. Then had to decide what I actually wanted to say. Still a work in progress, on purpose.
status: live-ishAt work: let mid-market teams set up projects, budgets, and costs in natural language. 0 → 1, still unfolding.
status: buildingNew places, new food, getting a little lost on purpose.
wanderlust, basicallyGenuinely one of my favorite things. Strangers become stories.
tell me everythingThe satisfaction of made that.
ship it!I like to move, compete, play.
game onThe fastest way to feel alive.
why not?Can't start my day without it. Non-negotiable.
one more cupMy recent love — what better way to learn than from history?
tell me the story— updated whenever something new happens, so check back often