i'm 16 and i live in kanpur. some weeks i disappear to delhi or bangalore when the work is louder than school.
i fell in love with my first computer in class four, literally, the way other kids fall in love with a sport. the next year and a half i spent making crude games in scratch out of an amateur cs book, the kind you play at dinner and refuse to shut down. my mother thought i slept at regular hours. fifteen days later i walked into the school techno fair with a spaceship game i could not stop blowing up, competing with class eight, and came home with first prize. i still do not really know what happened in those days, only that something in me decided this was the thing.
html in class five, python after that, java by class eight. covid lockdown was class six for me, and while a lot of people were bored i sat in a room coding until my eyes burned. by class eight i had a fitness tracker on the play store, politely reviewed by my parents, which nobody else had the patience to try. most of those early projects died quietly once nobody needed them, which is probably the correct fate.
class ten, 2024. i watched my grandparents fall in love with facebook reels until they needed a cab or groceries, and then the phone became this object they were angry at. i built orca so they could talk to whatsapp in normal language and pull the daily problems out of the glass. that season was competitions, cold emails to researchers, and the teenage need to prove i existed. some of it was hunger and some of it was vanity, and i am not ashamed of either. orca taught me i could build something with actual teeth. once i tasted that, i wanted my own problem again.
class eleven was jee. when a problem resisted my mind i did not want another chatbot or a forty minute lecture that only half addressed the stuck point. i wanted a diagram that moved, the step that makes a synapse fire, the step no teacher had been able to produce the night i wrestled electrostatics. i left the jee track. if the video did not exist at the price and quality a student could afford, i would build the machine that made it, mostly because i was the student who needed it. that machine became vivacity.
last summer i boarded a plane to bangalore with friends and a half-serious bag of clothes because staying in kanpur felt like hiding. we had two proud days sleeping at a shell petrol pump with laptops on our laps after a hackathon, waiting for prize money to clear. it was ridiculous and i would do it again. with help from pratham pengoria i met ravneet singh of prolearn, who let three rowdy teenagers waste his time after security took exception to us. he roasted our thinking, then hired me onto their video pipeline when we came back with demos. i work with him on stipend and credit, which is not enough to offset engine costs and is still heartening. i work with aditya bhatia and tanish anand, and i am the contact person for most of this.
hard exams are still good for serious children even when they feel cruel in the moment. indian families already invest years in jee and neet. the fashionable argument that these exams are meaningless is one i have always resisted. the pressure is unhealthy and coaching is predatory, and difficult problems are still a real filter for the kinds of minds that want to do physics and math. what gets me up is simpler: a lot of people need those filters and do not have access to the kind of explanation that makes the next step visible, especially on the specific problem that has them stuck.
a lot of money in ai video went toward looking like a movie. photoreal rooms, cinematic light, minutes of waiting for a clip. when an agent reasons i want the visual to arrive the way text arrives, in the same tab, at the speed of the model, frame by frame as the thought happens. conventional video generation paints heavy rgb grids on industrial clusters, which is slow in a browser and slow next to a language model. the thing i kept turning over was treating a visual the way a browser treats a page: a stream of light primitives and scene instructions the client draws locally. webgpu in the tab, skip the frames that have not changed, keep the model small enough for the memory a chrome tab actually has. coordinated diagrams, narrated steps, spatial reasoning made visible.
vivacity started as that hunger for moving explanations, cheap enough to call, exact enough to trust, in english, hindi, and hinglish, because that is how this country actually speaks. the research went further down. the thing we are building now is an executable world, where an agent can inspect state, act, simulate, fork a few futures, verify, and commit. the orbital workbench on this site is a local example of that loop, running in your browser, with no production api hiding behind it. just the verbs.