I was born in Casablanca in August 2002, and that's where I did most of my growing up before Munich entered the picture. When I was younger, I was fascinated by how the computer made decisions in single-player games — we called that kind of intelligence "AI" back then. I had a programming course at school where I learned Scratch, that block-based programming language.
By the time I got to TUM, I didn't have much coding experience. I struggled in the first semesters because of it, but adapted and picked up pace. I understood quickly that this was what I wanted to do, even while reading through thousands of coursework slides in my fourth language during my first and second year in the country.
In my first year of the bachelor's, I worked on a toy search engine as part of our semester-long project, with weekly submissions in the infamous PGdP (Praktikum Grundlagen der Programmierung). Ask any TUM student, and you'll see how many people's blood pressure spikes at that name.
Fast forward a few years: LLMs and generative AI tools like ChatGPT hit the public, and that reignited the same fascination I'd had with how that AI boss in that one video game thought and moved. I used my experience from the search engine practical to get a spot in a seminar on LLMs, back when they were still treated as "premium search engines" in the first weeks after release. I made a good impression there and turned it into a paper. I had fun testing the different LLMs of that time (Google Bard, GPT-3.5, GPT-4, GitHub Copilot) and seeing how they reasoned and arrived at answers.
Then came my practical course at Fraunhofer, which was deep into RAG, a hot topic at the time. By the time I was at Allianz, agents had become the hot topic, and I was tasked with testing and experimenting with agentic workflows. That's when it clicked: making systems out of agents was so much fun. Which agent specializes in what, how they interact, what they send to whom, how they break — it gave me endless things to think through. It snowballed from there through the TUM × SAP practical course and now BCG Platinion, to where I am today.
I'm still a student. I don't have an "AI Architect" title, and I'm not going to pretend I do. But if you put me in a room and asked whether I'd rather debug a function or redesign how ten agents coordinate without a central controller, I'd pick the second one every time. That instinct is what the orchestrator-free agent swarm I built at TUM × SAP came out of, made because the "one boss agent" pattern felt like the wrong shape for the problem.
Morocco doesn't get talked about much in the European AI conversation, and Munich's tech scene doesn't talk about Morocco much either. I'd like that to change a little, one project and one post at a time, not because either place needs a mascot, but because the gap between where I'm from and where I'm building doesn't feel like it should be as wide as it looks from the outside.
If you're reading this from Casablanca, or anywhere else that doesn't usually get mentioned in the same sentence as "AI research," I mean it literally when I say the gap is smaller than it looks. Most of what got me here was just showing up to the next thing, in a fourth language, without knowing yet if I was any good at it.