I find where products break, then I fix them.
Before business school, I spent two years at PhonePe working on insurance operations. I wasn't a PM, but I sat close enough to the product to see how things actually broke: funnels that leaked revenue, catalogues full of bad data, support queues that told you more than any dashboard could. Last summer I interned at Mastercard, where I built an AI tool that turned a 2.5-hour analyst workflow into a 15-minute one. Now I'm finishing my MBA at IIT Kanpur and looking for a product role.
Where I've worked.
A few things I keep coming back to.
The support queue knows more than the strategy deck.
At PhonePe I spent months running SQL queries to understand why users dropped off. Then one day I sat with the CX team and listened to actual support calls. I learned more in that afternoon than I had in weeks of data analysis. The data tells you what happened. The person on the phone tells you why.
Most growth ideas are broken funnels in disguise.
The motor catalogue at PhonePe had a 10% error rate on 2W listings. Nobody was asking for a new feature to fix it. But once we cleaned the data, conversion improved without changing a single screen. Sometimes the highest-impact work is the least interesting to talk about in a meeting.
An AI output without a confidence signal is just a guess.
When I built SHARP at Mastercard, the speed was the obvious win: 2.5 hours down to 15 minutes. But the thing stakeholders actually cared about was the scoring rubric. They wanted to know which answers to trust and which ones to verify. I think about that a lot when I see AI products that present every output with equal confidence.
If you're the user, you can't fake the problem.
I built StockHealth because I was frustrated with my own stock alerts. I built VibePath because planning a weekend trip from Kanpur was genuinely annoying. Working on something where you feel the problem yourself is different from working on something where you have to imagine the problem. The questions you ask are sharper.
Thinking in public.
Build logs, teardowns, and observations. Not polished thought pieces. Just things I noticed while doing the work.
Where PhonePe Insurance loses people
I spent two years watching the insurance purchase funnel from the inside. Here are the three drop-off points that hurt the most, and what I would change first.
Ops people make underrated PMs
Two years in insurance operations, then a summer building AI tools. Some thoughts on why people who've worked close to the mess tend to build differently.
Side projects.
All of these started because something annoyed me enough to build the fix.
Planning a weekend trip from a Tier 2 city in India means juggling Google Maps, MakeMyTrip, blog posts, and WhatsApp groups at the same time. VibePath takes your budget, where you're starting from, and what kind of trip you want, and gives you three fully costed itineraries across 62 destinations. Budget is a slider, not a filter. A second LLM fact-checks the first one's suggestions before you see them.
I got tired of stock alerts that told me a stock moved 3% without telling me whether that mattered for my actual portfolio. StockHealth sends me Telegram briefs three times a day, sorted by rupee impact instead of percentage. It has a conviction score per holding (1 to 10) that only updates when there's evidence across seven data layers: fundamentals, sector benchmarks, news, earnings, technicals, memory, and portfolio impact. Runs entirely on free-tier APIs.
Every dua app I tried either had no source verification or buried it behind ads. DuaVault is a simple archive where every dua links to its authentic source. I built it mostly for myself, but it clarified something I keep thinking about: in religious content, people won't use your product if they can't verify what it's telling them.
What I'm doing.
Reading, watching.
Read it once before PhonePe and thought it was good advice. Read it again after, and realized half the things I'd figured out at work already had names in this book.
Formula 1 was a niche European sport for decades, and then Netflix turned it into a global thing without losing the existing audience. That's a hard thing to pull off.
I read Ghalib because I like Urdu poetry, not because it makes me a better PM. Though it does remind me that most things worth saying resist being said clearly on the first try.