It started
with Dota 2.

I ended up at IIUI. It wasn’t known for its CS program, and there weren’t many resources for what I wanted to learn.

But I loved computer science. I loved the idea that I could make things.

I never actually made a game, by the way.

Watching it play was just… wow.

I knew this game. Seeing AI do that made it feel very real. I remember thinking, “I have to go into AI.”

Andrew Ng. Coursera. Yann LeCun’s deep learning course. Whatever I could find.

I didn’t have much around me at university for AI, so I did the studying myself.

I couldn’t get an internship in my second year. Third year wasn’t going much better.

Then, just before the summer ended, a mutual connection made an introduction. I joined a tiny startup: three people doing AI services.

I learned a lot there.

Six or seven times, to Red Buffer. The two Google Developer Experts in AI I knew of in Pakistan both worked there. I really wanted to work with them.

They finally hired me as a junior ML engineer in my last semester. I skipped my classes and went to my full-time job.

I was working on pose estimation at Red Buffer and watching an NVIDIA GTC talk. “Iqbal et al.” kept coming up.

Iqbal is a common Pakistani name. I looked him up: Umar Iqbal. I sent him a message, and he wrote back in detail.

That got me interested in research. It was the first time I’d heard of CVPR.

There were about 400 people in the Discord when I joined. Researchers, study groups, conversations I hadn’t had access to before.

Robert Joseph helped me give my first talk to a research community. It was about OpenAI’s CLIP.

Then I applied to be a community lead. I became the first regional lead for Asia.

I learned to reach out to researchers and invite them to speak. People from leading universities said yes.

We ran study groups, too. After attending Oxford’s ML Summer School, I wanted to help make something like it available for free.

Our summer school got over 2,000 registrations. More than 1,300 people joined live.

Through the Discord, I joined the Data Provenance Initiative. That led to papers at NeurIPS 2024 and ICLR 2025.

I also worked on a first-author paper about vision-language models. It was accepted at a CVPR 2025 workshop.

All this time, I was working full-time: two years at Red Buffer, then about two and a half at Roll.ai.

I was applying to grad school through all of this. For four years, every school rejected me.

The jobs, the research, the community work—that was happening while I was still trying to get in.