I Build Intelligent Systems That Hold Up Where Failure Is Costly

Before the Vocabulary Existed

At fourteen, I built “I System”, a speech interface that adapted to each person using it. Where the speech-to-text engines of the day were rigid and hard-coded, it gave every user a profile, learned their speech patterns through a reinforcement-style feedback loop, and stored their corrections in a SQL database so it could act on commands unique to them.

It won at Intel ISEF Egypt in 2011, and a year later I followed it with TRON OS. The instinct behind both, systems that adapt to people rather than the other way round, still drives my work today.

Where It Took Root

A fascination with how things work, and more importantly how they fail, set the direction for my undergraduate years at the German University in Cairo. I graduated as class valedictorian with highest honors, on a full tuition scholarship and with multiple academic excellence awards, then stayed on to teach.

The Turning Point

My six-month bachelor's thesis placement at the University of Stuttgart was my first serious encounter with computational modeling: building software to simulate how complex systems behave. The most interesting questions, I found, weren't about individual components. They were about uncertainty, interdependence and prediction. The instinct I'd had at fourteen finally had a discipline behind it.

A PhD, and a Commitment to Applied AI

I moved to Canada for a PhD at McMaster University, and from 2022 to 2025 held the Vanier Canada Graduate Scholarship, one of the country's most prestigious doctoral fellowships. What began as research into ML-driven risk modeling became a commitment to applied AI in its own right: how do you build systems that learn reliably, generalize under distribution shift, scale to production and stay interpretable when the stakes are high?

I built production TFX pipelines on Google Cloud, developed graph-based models of cascading failure across interdependent networks, and published 8+ peer-reviewed papers. The hardest open problems in AI, I came to see, aren't about accuracy. They're about reliability, oversight, and what happens beyond the training distribution.