I build software and AI for people who need it.
I write code, learn languages, break things and fix them, then start the next project sure it will be easier. It rarely is, which is the most fun haha.
Late afternoon
Software and AI with a person on the other end.
The largest share of my work is medical AI research with LiGHT Lab, the Laboratory for intelligent Global Health and Humanitarian Technologies at EPFL, which works in close partnership with the Harvard T.H. Chan School of Public Health. I joined its core team in August 2025 as a research student assistant. The work is research and software engineering at once: evaluating medical language models, retrieval across data of every kind, and the infrastructure underneath both.
I came to Data Science and Artificial Intelligence at SUPSI by way of Computer Science at EPFL. Morocco is home. What I like about Switzerland is the mix of people, and how differently they come at the same problem.
Golden hour
Medical language models, and the means to test them.
These are the projects my time at the lab has gone into. The paper further down came out of the first one.
MOOVE is where the lab finds out whether a medical language model is safe to rely on: an adaptive trial in which clinicians across many countries judge the models on clinical cases. I am part of the team behind the platform itself. I helped move it from Firebase to Convex, simplify and document the schema, build the contribution and review workflows the clinicians use, and turn their judgements into preference datasets for training. The project lives at jointhemoove.org.
Meditron is the lab's medical language model. My part was the data: I curated the datasets it learns from and ran the evaluation jobs on the CSCS Alps and EPFL RCP clusters. The open release is on GitHub.
MMIRAGE automates reformatting and augmenting datasets with language and vision-language models. I contributed to it, and the source is on GitHub.
mmore is a multimodal RAG pipeline that indexes PDFs, office files, audio, video and web pages across many machines. I contributed to it, and version 2 shipped in July 2026. The source is on GitHub.
Macharia P, Kachimanga C, Mahende MK, and others, including Azgaoui OZ. Evaluating large language models as clinical decision support tools in primary healthcare settings: protocol for a multi-country comparative validation study on expert-adjudicated hypothetical vignettes (hypMOOVE-PHC). medRxiv, 15 September 2026.
Read the preprint or see Google Scholar.
Written on LinkedIn by people I had the chance to work closely with at the lab: my supervising professor and a senior colleague.
Omar, or Azo as we know him at LiGHT, is one of the most capable, focused, and impressive undergraduates I have worked with (top 1%). He learns quickly, looks beyond the immediate task, and has a strong instinct for identifying what a team actually needs. He brings forward thoughtful ideas, takes ownership of them, and follows through with work that is useful and well executed.
He is also a superb colleague: curious, dependable, receptive to feedback, and confident enough to contribute meaningfully alongside senior engineers and researchers. His humour, warmth, and professionalism have made him a truly valued member of LiGHT.
I remain available to share even more effusive praise for Azo as required.
Omar has been a fantastic addition to our lab this year, managing to bring his ideas to help the team progress further, making tools to make our lives easier and he got things done (and well done) for the MOOVE and some of my projects within the lab. He is quite sociable, a great addition that makes the lab even nicer of a workplace, while still showing true professionalism.
Sunset
Most of it started with someone next to me needing it.
Sometimes a reviewer, sometimes a classmate or a lab mate. These are the ones that stuck.
A GitHub Action you request like any other reviewer. It reads the whole repository before it says anything, leaves inline comments with suggestion blocks, gives a verdict, and never approves, kinda like academia's Reviewer #2. hihi
Source on GitHub, or install it from the Marketplace.
The DTI timetable as a calendar subscription. Pick your class, drop the lessons you do not follow, and your phone stays in sync every time the timetable is republished.
A deadline tracker for research conferences. The lab needed one, so I built it: countdowns to submissions, a timezone-aware calendar, ICS export and Slack reminders.
The trips from EPFL to Gruyères, with the stops and changes of one of them.
A route planner for Swiss public transport, built with a classmate in 2025 on the connection scan algorithm, with multi-criteria routing and a JavaFX interface.
The town, with its gym and its Pokémon Center.
A turn-based creature game in Java that a classmate and I built in 2024, in the spirit of Pokémon. It was my first real object-oriented design.
Blue hour
Where most of my evenings went.
AGEPolySport team, since 2025
Sport got me in the door, and the people kept me coming back.
1 of 7: About twenty players lined up on the podium steps of a sports hall, under a banner of a red and white owl.
EPFL AI TeamML engineer, 2025 to 2026
I found people here who build things for the fun of it, and I fit right in.
1 of 4: About fifteen students posing together in front of an EPFL AI Team stand under a concrete ceiling.
Students 4 StudentsTeaching assistant and event staff, 2025 to 2026
Every year a new group arrives not knowing where anything is, and for one week I get to be the person who knows, and who shows them around.
1 of 3: Students in aprons and lanyards carrying trays of bread through a canteen, one smiling at the camera.
Staff, 2026
A faculty is a lot of people who rarely sit at the same table. For one evening, we made sure they did.
1 of 3: Long rows of dinner tables full of students under the glass roof of an atrium; an inset shows three of the staff.
Forum EPFLVolunteer, 2025
I like the hours before a big event opens, when everything still has to come together and nobody is watching yet.
1 of 2: A green and black Formula 1 car on a blue carpet at a fair stand; an inset shows a student in a suit and glasses.
DESMNUDelegate, then trainer, 2019 to 2023
I learned to speak to a full room here, and later how to help others do the same.
1 of 13: Four Model UN delegates on stage, Omar in a blue suit holding a glass award
Dusk
The rest of the day.
I grew up in Rabat and came to Switzerland to study, and a few trips have taken me over a border or two since. The list of places I want to see grows faster than the one I have been to.
Been to
Want to see
Last updated: October 2026
This is a personal portfolio website created by Azgaoui Omar Ziyad. Its sole purpose is to showcase my projects, skills, and professional experiences. No personal data is collected for commercial or advertising purposes.
The cookies used on this site are strictly for:
By continuing to browse this site, you accept the use of these cookies.
Used to analyze site traffic (pages viewed, visit duration, device type, approximate location). Data is anonymized and used solely for statistical purposes. View Google’s Privacy Policy.
Used to analyze visitor interactions (clicks, scrolling, mouse movements) in order to improve navigation. Data is also anonymized. View Microsoft’s Privacy Statement.
Used to look up which organization the network behind a visit belongs to, such as a university, a company or an internet provider. Your IP address is sent to ipapi.is for that lookup alone, and only the organization’s name is passed on to the tools above. View the ipapi.is Privacy Policy.
For any inquiries, please contact me at:
omarziyad.azgaoui2005@gmail.com