Capt Venkat 🇮🇳

My name is Arjun Shah, a 14-year old entrepreneur, and this is my story. 👇 I was seven years old when I first discovered coding. I started with Tynker, solving little puzzles, and eventually moved on to Python. I loved the idea that I could create something entirely from scratch, just using a computer and my imagination. But there was one problem. I HATED syntax. So I quit. But that spark inside me never died. When I was 12, I discovered Cursor, and it changed my life. I used it to build a crappy AI-powered plant identifier in HTML+CSS (I didn't even know what Github was), but it really showed me that I could use AI to build something real. I started experimenting with automations, building applications, and exploring neural networks. I became obsessed with 3Blue1Brown and Karpathy, and started building my own SaaS's as well. At thirteen, I built therooted.ai, a startup that went on to win the Stanford Graduate School of Business LISA startup competition. I was the youngest ever winner. But I was just getting started. At fourteen, I started putting myself out there. I participated in Chamath Palihapitiya's 8090 Hackathon, where I reached the podium. Recently, I placed second (800+ participants) at the Composio and Nebius Builders Ship Hackathon (in SF, on a yacht!!!). I got to present my work at a real convention in SF, and met so many cool people! I also received twenty thousand dollars in E2B credits to support my development work. Around this time, I became fascinated by a bigger question: what if AI could do more than help someone write code? What if it could independently build an entire software product? That's what led me to create Loopy (loopy.yachts), an autonomous software engineer. You give it an idea, and it researches the problem, plans the implementation, writes code, runs tests, fixes mistakes, and keeps iterating. [this is what i used at the yachtathon] But as I worked on Loopy, I discovered a problem that almost every long-running AI agent faces. Context. An agent might work perfectly for five minutes, but after five hours, it can start forgetting decisions, repeating the same work, and consuming enormous amounts of tokens just to remember what happened earlier. So I decided to build something to solve it. That became SuperCompress.dev, an open-source context compression engine designed to help AI agents remember what's important while discarding unnecessary information. I've racked up over 5 million views since launch, and we're about to hit a billion (with a B!!!) processed tokens. But I still feel like I'm at the very beginning. Because for me, this was never really about learning how to code. It was about learning how to build. And this is only the beginning of my story. Follow along to see how it goes.