Computer Science & Machine Learning. I build systems that work where most technology gives up. IoT sensors in villages with no internet. Machine learning on phones people already own. The problems that matter are rarely the problems that are easy.
I want to be direct with you, because I think you deserve that more than you deserve comfort. Most students in their first semester are still figuring out where the library is. I filed a patent. Not because I am unusually gifted, but because I became genuinely curious about a problem that almost no one in my environment was trying to solve.
The system is called Project CHIRP. It uses ESP32 sensors and on-device computer vision to detect crop anomalies in real time, then delivers alerts via SMS to feature phones. No cloud. No smartphone required. No one told me to build it. No course required it. I simply noticed that most agricultural technology assumes farmers have smartphones and stable internet, which is a bit like assuming everyone has a swimming pool. It is not how the world actually works.
Today I am a Machine Learning Intern at BetterRoads, building pothole detection from smartphone accelerometer and gyroscope data. I own the signal preprocessing and classification pipeline. The challenge is not the algorithm. The challenge is building something that works on real roads, in real cars, with real noise, across devices that cost two hundred dollars and devices that cost a thousand. Theory is clean. Reality is not.
I also built Grafite, an exam prep platform that grew to 1,000+ students with zero paid marketing. Spoiler: if a product is actually useful, people find it. If they do not find it, the product is probably not as useful as you think.
Building a mobile pothole detection system using smartphone gyroscope and accelerometer data. I own the preprocessing pipeline and the signal classification logic that distinguishes road anomalies from normal driving vibrations.
The real work is not the model. The real work is understanding that a pothole in Hyderabad and a pothole in Mumbai produce different signatures, that cheap phones distort accelerometer readings differently than expensive ones, and that most of what machine learning literature calls "noise" is actually the signal you should have been listening to all along.
Designed and built an end-to-end IoT system for smallholder farmers. A distributed wireless sensor network using ESP32-NOW with on-device HSV color analysis and frame differencing for real-time crop anomaly detection.
No cloud dependency. Alerts delivered via SMS to feature phones. Built for the environments where most agricultural technology fails, which is to say, built for the world as it actually exists rather than the world as Silicon Valley imagines it.
An IoT agricultural decision support system. Combines hardware sensors, computer vision, and weather data. Eliminates cloud dependency for low-connectivity rural environments by delivering alerts via SMS to feature phones. It works precisely where most technology stops working, and that is not an accident. It is a design choice.
A competitive exam prep platform for JEE, NEET, and BITSAT aspirants. Grew to 1,000+ active students through organic discovery. Co-founded a Discord community of 1,000+ members for peer support and doubt resolution. No marketing budget. No growth hacks. Just a product that solved a problem students actually had.
The only clean, self-contained Flatpak Spotify patcher for Linux. Reverse-engineered Spotify's xpui.spa bundle to apply surgical string and regex patches to compiled JS/CSS assets. Eliminates audio, video, and banner ads. Twenty stars on GitHub with zero promotion, which is either modest or precisely the point.
An open-source tool automating government portal integration via AI-driven web scraping. Achieved 95%+ form field detection accuracy with a self-healing code generation loop and pattern library for knowledge retention. Because government websites change constantly, and brittle automation is worse than no automation at all.
I am always open to interesting problems, hard engineering challenges, and good conversations. The world does not need more software. It needs more software that understands the people who use it. If you are working on the second kind, I would genuinely like to hear about it.