I'm 19, studying Data Science at IIT Madras. I own the code for a live consumer voice product that has never paid for a single user. Most of what I know came from the projects that didn't work.
That sounds obvious. It is not how most software gets made. Most of it gets made by deciding what people want, building all of it, and finding out at the end.
I've shipped in healthcare, privacy, education and consumer voice. Four unrelated domains, one repeated lesson: the technical problem is almost never the hard part. Knowing which problem you are actually solving is the hard part.
So the skill worth compounding isn't writing code faster. It's getting to the truth sooner.
I own the codebase end to end: voice onboarding, the matching pipeline, the web app. Past 200 members and 50,000 organic views without paying for a single one of them. A previous technical lead had left, I took a one off audit task, and I have been the person responsible ever since.
Vector similarity narrows the field, then a reasoning model ranks what's left and writes down why. The written rationale is the part that matters. A recommendation nobody can interrogate is just an assertion.
Where multi step agent systems actually hold up and where they quietly fall apart. Publication in progress.
The products I keep studying are the ones that look inevitable in hindsight. None of them looked inevitable while they were being built. They looked small.
The last one is the question I ask most now. Anything that gives people a judgment about their own life has to earn that right, and the bill for being wrong is paid by them, not by me.
Took a one off audit task on a live consumer product. Rebuilt most of it and became the person who owns it.
Voice onboarding, the matching pipeline and the web app all ship from my hands.
Past 200 members, a sold out paid launch event, zero paid acquisition.
1st place at the Google Developer Groups Vertex AI Hackathon, from a field of roughly 2,400 teams.
Top 12 at the world's largest civic technology hackathon, 250,000 participants.
2nd runner up twice more: IIT Delhi Innov8, and Agentic Among Us with Eightfold AI.
Shipped production features used by 47,000 students at IIT Madras.
Started agentic AI research with IIT Madras. Publication in progress.
Began the BS in Data Science and Applications at IIT Madras.
Moved from things that got graded to things that got used.
Started writing software at 15, not because I liked code, but because it was the cheapest way to find out whether an idea was real.
Finishing the automated matching and feedback loop on the voice product. New signups are paused until it's done, because a waiting list nobody is matched from is worse than no waiting list.
Diploma stage at IIT Madras: databases, machine learning foundations, application development.
Agentic AI research with IIT Madras, publication in progress.
Trying to build a working rhythm that doesn't depend on late nights. This one is going less well than the others.
A previous technical lead had left. I did a trial audit task to see how the system actually worked. I ended up owning it.
2,400 teams competed. Most built something technically impressive. We won by building something simple and spending the saved time on the parts that were actually getting judged.