About me

I've been building AI systems for about eight years now. I started in NLP — Arabic intent classification, dialect ID, the kind of problems where the data is messy and you have to fight for every F1 point. From there I drifted into computer vision (real-time proctoring, in-store CCTV analytics) and these days I spend most of my time on LLMs, RAG, and VLMs.

What I enjoy is the full loop — research a model, ship it, run the infra it lives on, and then watch real users actually use the thing. I like the parts most ML engineers don't love: the GStreamer plugin, the pybind11 binding, the KServe config, the Temporal workflow that retries until it works. The model is usually the easy bit.

I'm currently leading AI at Search Capital, turning a dormant candidate database into the platform's primary matching engine — bidirectional matching, explainable scoring, and a self-hosted stack we fully control.