Siddhartha grew up in India and, after running the gauntlet of the national entrance exams, studied computer science at IIT Kanpur. His research there had a cinematic quality to it: autonomous underwater vehicles that used computer vision and machine learning to navigate on their own. He published at international conferences and spent an exchange semester at Aalto University in Finland. 

The through-line, even then, was perception. He was teaching machines to make sense of an environment they could not be told about in advance.

But pure research moved slowly, and Siddhartha wanted a faster signal. Startups had shorter feedback cycles, real users telling you in days what a paper might take years to prove. He moved to Montreal to work as a machine learning researcher, then left to build. His co-founders were two batchmates from IIT Kanpur, Pratyush Rai and Sirsendu Sarkar, friends from a class where their roll numbers had sat almost side by side. He was trading long research horizons for the fastest possible signal. He was about to get more of it than he bargained for.

Merlin launched just as ChatGPT was breaking through to the mainstream. It began as a browser extension that could summon AI onto any website with a single keystroke, a spotlight for intelligence wherever you happened to be working. Then a popular AI YouTuber dropped it into a top-five-tools video, and the signups arrived in a flood, so fast that the team was holding the whole thing together on top of OpenAI APIs that kept buckling under their own demand.

From there Merlin grew from a one-shortcut extension into a full web suite, with multiple models under a single subscription, the whole premise being that you should not have to marry one ecosystem to get the best of what AI can do. It became one of the largest AI extensions to come out of India, with more than 20 million users worldwide. The researcher who wanted faster feedback had found it, at a scale that nearly broke the servers. 

Which brings us back to that thousand-message chat. 

The team kept noticing the same pattern: power users who never started a new thread, because starting over meant explaining themselves all over again. Siddhartha's read on it was almost philosophical. A conversation is the truest, most human record we produce, and also the most ephemeral. No document captures it. No tool holds onto it. It simply evaporates. So the vision for Thine wrote itself: capture all of that conversational context, so you never have to explain yourself to AI again. 

But how does it work?

Building the product is only half the job. Getting anyone to notice it is the other half, and Siddhartha's team learned that craft the long way. 

An early series called Tech Branch brought cartoon-style tech humor to the timeline and got some traction, but only by brute force, cross-posting into more than ten communities to squeeze out every bit of reach. The breakthrough came from a simpler instinct: take a format people already love and drop it somewhere it does not belong. 

They had watched people joke on Twitter about tech culture in the voice of The Office, with no one actually making the thing. So they made it. The script took three or four days, and they knew before they hit post that it was, in their own words, viral worthy. It cleared sixty thousand views fast, then two hundred thousand, then three hundred thousand and beyond. The team behind Sora reposted it. And it spawned a meme that now travels on its own, which is the real prize. It became part of the culture. 

The thesis underneath is pure Siddhartha: juxtaposition stops the scroll. Put two formats from two different worlds next to each other, and the collision is what makes a person stop moving their thumb. He has a favorite way of saying it, borrowed from the 2017 machine learning paper that started this entire era. Attention is all you need. It was true for transformers, and it is still true for growth.

Ask Siddhartha for the lesson of the whole journey and he does not reach for a growth hack. 

He reaches for two words: patience and ambition. 

Every product, every feature, every content experiment that flopped or half-worked, all of it compounded into the next step. The underwater robots taught him perception. Merlin taught him distribution. The series that only sort of worked taught him exactly what the one that exploded would need. Overnight success, he will tell you, always has a long and invisible history behind it. 

Siddhartha Saxena has spent a decade teaching machines to understand context, from the ocean floor to the browser to the conversation, and he has done it with the patience of a researcher and the ambition of a founder. 

Making “explaining yourself” obsolete isn’t a half bad way to describe the entire project.

The AI Collective is built by volunteers across 180+ chapters in 40 countries.

Thank you to the thousands of volunteers around the world who make this work possible. We truly could not do this without you.