Somehow, I keep finding things before they have a name.

The moment

In 2019, I came across a video from OpenAI: AI agents teaching themselves to play hide and seek. I watched them learn the game. Then I watched them break it.

They found the edge cases in the game's physics and used them, doing things the people who built it never intended. No code is perfect, and they had found the cracks on their own. That was the moment AI stopped being a tool to me and became the thing I wanted to understand.

Before that

I grew up in a Bengali family that gave me every chance it could reach: art school, music, karate, and a computer from class 4. My elder brother did the rest. When I was nine or ten he was explaining the fourth dimension to me, and relativity, and how big the universe is. I spent years chasing those ideas through YouTube videos and blog posts I was too young for.

From class 6 I lived at Jawahar Navodaya Vidyalaya, a residential school where you learn to look after yourself and to talk to everyone. I went to science congresses, reached the nationals in drama, played chess at the regional level at fourteen, and was a finalist at Toycathon. The friends from those six years are still the closest I have.

Two hundred videos

Without a proper device for most of school, I played games, rooted phones and started YouTube channels. I made more than two hundred voiceover videos. Nobody saw my face, but I learned how to edit so people stay, how to package a title and a thumbnail, and how an algorithm decides who gets seen. I didn't know it then, but that was my first lesson in distribution.

Early access

In 2022 I got into the DALL·E beta, back when most people didn't know the company's name, and started making things with it. When ChatGPT arrived at the end of that year I was an early adopter, like I always was with anything new. But I went off to the part that actually pulled me: the creative, generative side.

Making it mine

I trained full models with DreamBooth on borrowed Kaggle and Colab GPUs. When LoRA arrived, I trained LoRAs. After my boards, my brother put me through 100xEngineers, and I moved from generating images to engineering the systems behind them: Automatic1111, then ComfyUI, then reusable node-based workflows deployed for serverless inference, putting that capability into clients' hands through Visualsverse and Avatar GenZ.

Those graphs covered virtual try-on, character sheets, coherent movie sequences and specialised systems like TextureFlow. The challenge was never one good image. It was making intricate pipelines reproducible, deployable, and simple enough to become products.

The films

Around the same time I was making short films with generative AI, and not the kind where the AI does it for you. The models were bad, so I directed every frame by hand. One was a biography for Tridib Nandy, about what happened to him in police custody. Another went to Project Odyssey and came back with prizes. I also ran social media for a cannabis venture firm, which taught me more about attention than I expected.

Systems that deliver

In December 2024 I started learning n8n, and it was hard. My brother taught me to think in systems. In January I joined Cash Cow Labs, the company he started with Sharique, who became my second mentor. When it joined forces with BenAI, I was one of the first six people there.

I came in as a junior automation engineer: the automations behind Ben's videos, custom software for enterprise clients, our products deployed on their own servers. I was a forward-deployed engineer before anyone called it that. By the end I was leading engineering, delegating to a team, and running the custom build for our biggest client, which shipped fifteen to twenty thousand content pieces a week at its peak. The company went from nothing to about a million in ARR in a year. I was part of that, not the reason for it.

We never automated anything we didn't understand. We sat with the client, paid people who were the best at the work to show us how they did it by hand, ran it by hand ourselves, and only then built the system, with a person kept exactly where judgment was needed.

A pause

When we all left BenAI together, I took a long break. I travelled through Thailand and Vietnam with my brothers, sat with meditation and deep yoga, and came back seeing my own work from somewhere further away. Along the way I'd helped a few friends land their first jobs, and then watched them do the same for theirs. That is still the thing I'm proudest of.

Now

Now I am building Vibe Life. Right now that means distribution: content systems where AI carries the volume, from idea to edit to thumbnail, and a defined process keeps the quality high and the voice the founder's own. The vision is larger and I am still growing into it: a way to manage your digital self.