Prospective Students

Join the GenMI Lab at Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), Abu Dhabi, and build AI that medicine can rely on.

We work where machine learning meets medicine and biology: medical foundation models, vision–language reasoning, causal discovery, and personalized healthcare. We release what we build, we publish in the leading venues, and we care that the work holds up on real clinical and human data.

I welcome enquiries from prospective MSc and PhD students, postdoctoral researchers, visiting students, and remote research interns. This page explains what we do, what I look for, what you can expect in return, and what life is like here. Please read it before you write to me; it will make your email much stronger.

1) What we do at GenMI

GenMI is my research lab in Computational Biology at MBZUAI. Our aim is medical AI whose answers can be checked: grounded in the image, supported by evidence, and tested against real human data. Three lines of work run through the lab.

Medical foundation models

Models that bring images, language and other clinical data together for reasoning, spatial grounding and report generation.

MedMO, our multimodal medical model, is open on Hugging Face and was published at CVPR 2026.

Causal discovery

AI whose claims rest on explicit causal structure that can be queried, tested and revised when the evidence changes.

DoAtlas-2 generates its own research questions and tests them in human data covering more than 720,000 participants in 48 countries.

Personalized healthcare

Understanding individual variation, from population-scale biology to signals recorded from the body over time.

Our oral microbiome study was accepted in Nature Communications; the WLMs family covers ECG, continuous glucose monitoring and sleep.

What that looks like in practice

We ship. Models, code and project pages are released openly. See the DoAtlas and MedOS model families.
We publish where it counts. In 2026 the group announced papers at CVPR (four, including an oral), ICML, ACL, EMNLP, MICCAI and CIKM, and in Nature Communications, IEEE TIP and IEEE JBHI. Browse the publication list.
We help shape the field. We organize workshops at MICCAI: Mechanistic Interpretability for Medical Foundation Models, Agentic AI for Medicine, and SAFER.
We stay close to real use. I also lead MedOS (AI for medicine) and MedOmni (evidence-based medicine), so the question “would this work in practice?” is never far away.

Current students work on medical vision–language models, causal discovery in human health, adaptive multi-agent AI systems, health time series, mechanistic interpretability and clinical NLP. The wider group includes PhD students at UNSW Sydney and the University of Southampton. Meet the team →

2) Should you do a PhD?

A PhD is not more coursework. It is training to become an independent researcher. You will spend years on problems where the answer is not known, the path is unclear, and progress depends on your ability to think, read, build and iterate.

Strong reasons to do a PhD

You enjoy open-ended problem solving and deep thinking.
You want to publish and contribute new ideas to the field.
You want a research career, in academia or an industrial research lab.
You are excited by building systems that push the frontier in medical AI.
You are motivated by impact: better diagnosis, clinical workflows and patient outcomes.

Common wrong reasons

“I want Dr. before my name.”
“I’m not sure what to do next, so I’ll do a PhD.”
“A PhD guarantees a higher salary.” It does not.
“I want to delay joining industry.”
“I will work when I’m given tasks.” A PhD requires self-drive.

The best sign that a PhD is right for you: you can spend weeks on a difficult problem, fail repeatedly, learn from the failures, and still be curious enough to try again.

3) Is my group a good fit for you?

We work on personalized medicine and medical imaging, with an emphasis on multimodal AI that can understand images, reason clinically, ground its decisions spatially, and remain trustworthy under distribution shift.

Projects we love

Medical foundation models that unify tasks across radiology, pathology, ophthalmology, ultrasound and more.
Vision–language models for clinical reasoning: multi-step, evidence-based and uncertainty-aware.
Spatial grounding: localization, alignment, region-level reasoning and explainability.
Causal discovery in human health: finding and testing causal relationships in real cohort data.
Trustworthiness: hallucination reduction, robustness, calibration, interpretability and evaluation protocols.
Personalized and longitudinal healthcare AI: modeling trajectories over time, from clinic visits to wearable signals.
Agentic AI for medicine: multi-agent systems, retrieval and clinical NLP.

You will probably thrive here if…

You want your work to matter for patients, not only for a leaderboard.
You like building real systems and then testing them hard.
You are comfortable moving between ML, medicine and biology.
You want to aim for top venues and open releases.

We may not be the right match if…

Your main interest is software engineering without research.
You want purely theoretical work with no healthcare motivation.
You prefer to be told exactly what to do each week.

4) Skills you should have

You do not need everything below on day one. You do need the core, depth in at least one direction, and evidence that you learn fast.

The core (everyone)

Mathematics

Linear algebra, probability and statistics, and optimization, at the level where you can read a methods section and work out what a loss function is really doing.

Machine learning and deep learning

Training, debugging and evaluating neural networks. Transformers, self-supervised and multimodal learning, and a feel for why models fail to generalize.

Programming and engineering

Fluent Python and PyTorch. Git, Linux and the command line. Running and tracking experiments on GPU clusters, and writing code that someone else can rerun.

Experimental rigor

Strong baselines, ablations, clean data splits, more than one seed, and honest reporting. In medicine, a result that does not replicate is worse than no result.

Reading and writing

Read a paper critically and summarize it in a page: the claim, the evidence, the weakest point. Write clear technical English and make figures that explain themselves.

Communication and teamwork

Explain your work to ML researchers and to clinicians. Give useful feedback, take it well, and say early when you are stuck.

Depth in one direction

Experience in one of these is enough to start. The rest you will learn here.

Direction	Useful background
Medical foundation models and VLMs	Vision–language pretraining, fine-tuning large language models, detection, segmentation and grounding, large multimodal data pipelines, measuring hallucination and calibration, familiarity with medical imaging modalities.
Causal discovery and evidence	Causal inference (graphs, identification, estimation), biostatistics and study design, working with cohort data, knowledge graphs, LLM agents.
Physiological signals	Time-series and sequence modeling, signal processing, wearable and sensor data such as ECG, continuous glucose monitoring and sleep.
Computational biology	Omics or microbiome data analysis, statistics for high-dimensional biological data, bioinformatics tooling.
Agentic AI and clinical NLP	Large language models, retrieval-augmented generation, multi-agent systems and tool use, careful evaluation.
Trust and interpretability	Mechanistic interpretability, robustness under distribution shift, uncertainty estimation, fairness.

The skills that are hardest to teach

Ownership: you decide the next step without waiting to be asked.
Persistence: you keep going after the third failed experiment.
Honesty: you report what happened, not what you hoped for.
Speed: you get from idea to first result quickly, then refine.
Curiosity about medicine: you do not need a medical degree, but you must want to understand the clinical and biological side of your problem.

A self-check before you email

Can you reproduce the main result of a recent paper, from its code or from scratch?
Can you explain one of our papers, including where it is weakest?
Can you point to code you wrote that someone else can run?
Can you describe a project that failed, and what you learned from it?

If you can say yes to most of these, you are ready to apply. If not yet, they are a good plan for the next few months.

What I look for at each stage

MSc students and interns: a solid core and one substantial project you can walk me through in detail.
PhD students: the core, plus a completed piece of research (a thesis, a paper or a serious open-source contribution) and the beginnings of your own direction.
Postdoctoral researchers: a record of first-author work in strong venues, and the ability to lead a project and mentor students.

5) How to choose a PhD supervisor

Many applicants email dozens of faculty members with the same template. This usually fails. A PhD is a long collaboration, and compatibility matters as much as the topic.

What to check before emailing

Do you like the supervisor’s recent papers and research direction?
Are students publishing regularly in strong venues?
Does the group culture suit you: hands-on or hands-off, applied or theoretical, collaborative or independent?
Do you understand what you would likely work on in the first year?

The strongest emails show that you have read one or two papers, can explain why they matter, and can propose a direction you want to explore, even if it is preliminary.

6) How a PhD works, and what strong students do differently

A PhD is driven by the student. I will guide, challenge and support you, but you must own the research. The difference between struggling and thriving is usually habits and process, not intelligence.

Strong students typically…

Read consistently (papers, code, evaluations) and keep notes.
Implement quickly, test ideas, and iterate with evidence.
Write early. A living draft beats a perfect draft later.
Communicate clearly: what they tried, what failed, what they learned, what is next.
Build strong baselines and trust their experiments.

Students who struggle often…

Wait for tasks instead of defining next steps.
Spend too long preparing without building or testing.
Avoid writing until the end, which makes publishing hard.
Run experiments without careful evaluation or ablations.
Disappear for weeks, then compress their progress into one meeting.

7) What you can expect from me

My goal is to help you become a strong independent researcher who can publish in top venues and build medical AI systems that matter. In practical terms:

A clear starting direction: a reading list, baselines, milestones and a realistic plan for your first project.
Regular supervision: weekly meetings, more when needed, and active iteration on ideas.
Writing mentorship: structure, positioning, clarity and the craft of a top-tier paper.
A top-venue focus: NeurIPS, ICLR, ICML, CVPR, ACL, EMNLP, AAAI, WWW, MICCAI and leading journals. Quality over quantity.
Work that gets used: opportunities to contribute to open models and releases such as MedMO and DoAtlas.
A good research culture: high standards, fast iteration and supportive collaboration.

8) What I expect from you

A successful PhD takes more than intelligence. It takes consistency, discipline and ownership. If you join my group, I will expect:

Strong fundamentals: the core skills above, at a working level.
Practical ability: clean code, reproducible experiments and careful evaluation.
Independence: you learn what you do not know and come back with solutions.
Ambition: you aim for top venues and meaningful scientific contributions.
Integrity: honest reporting of results, including negative ones, and respect for research ethics and patient data.
Presence: you are part of the group, you help others, and you show up.

If you feel stuck or overwhelmed, tell me early. Silence is the biggest productivity killer in a PhD.

9) Life at MBZUAI

MBZUAI is a research university dedicated to artificial intelligence. Its campus is in Masdar City, Abu Dhabi, a walkable district that is also home to AI and technology companies including G42, Core42 and Siemens. The student community is small and international: 929 students from 71 countries in the 2026–2027 academic year.

On campus

Housing: furnished student apartments on campus, available to first-year graduate students.
Food: a canteen serving breakfast, lunch and dinner, plus restaurants and cafés.
Sport: gyms, a swimming pool, basketball, volleyball, badminton and table tennis, with intramural and intercollegiate teams.
Clubs and events: student clubs from chess and rock climbing to debate, a student government, hackathons, industry events and International Day.
Health and wellbeing: health insurance for all students, a first aid clinic open around the clock, and a confidential counseling service.
Everyday needs: prayer rooms, a park next to campus, a shopping mall nearby, a university shuttle, and help opening a bank account during orientation.

Research team taking a group selfie on a waterfront promenade at dusk. — Celebrating the release of DoAtlas-2 and welcoming our new member, Abdullah. More from Team Life →

In our group

We meet every week, work side by side, and make time for team lunches and celebrations when a paper is accepted or a model is released. New members are welcomed into the group from their first week.

Funding

MBZUAI states that funded graduate students receive free tuition, a monthly stipend, one return flight home each year, health insurance, and a university-sponsored UAE residence visa and Emirates ID where eligible.

Funding terms and stipend amounts are set by the university and can change. The PhD admissions and master’s admissions pages are the authority. More on student life, housing and campus facilities.

10) Living in Abu Dhabi and the UAE

Almost everyone at MBZUAI arrived from somewhere else. A few things that make settling in easier:

Safe

Abu Dhabi was ranked the world’s safest city in 2026 for the tenth year running in Numbeo’s index, as reported by the Abu Dhabi Media Office.

International

Most residents of the UAE come from somewhere else, so being new is normal. English is used widely in daily life, and you will find food, communities and places of worship from all over the world.

Well connected

Masdar City is next to Zayed International Airport, with direct flights across Asia, Europe and Africa. Dubai is about an hour away by road.

Plenty to do

The Corniche and the beaches, the desert and the mangroves, the Sheikh Zayed Grand Mosque and Louvre Abu Dhabi, and a year-round calendar of sport and culture.

Worth knowing before you come

Climate: winters are mild and sunny, and are the best time to be outdoors. Summers are very hot and humid, and life moves indoors.
Culture: the UAE is welcoming and also has its own laws and customs. Respect for them, especially during Ramadan, goes a long way.
Getting around: Masdar City is walkable, taxis are easy to find, and the university runs a shuttle to shopping malls and other key destinations.
Paperwork: for funded students the university sponsors the residence visa and Emirates ID where eligible, and orientation covers practical steps such as opening a bank account.

11) How to apply

There are two parts, and you need both. Emailing me is how we discuss research fit. The formal application to MBZUAI is how you are admitted. One does not replace the other.

Step 1: contact me about research fit

Email imran.razzak@mbzuai.ac.ae with a subject such as PhD Application — [Your Name] — Medical VLM / Foundation Models. Please include, as PDFs or links:

CV: GPA and rank where relevant, awards, publications and research experience.
Transcripts: unofficial copies are fine for a first contact.
A short research proposal (1–2 pages): the problem, why it matters, your idea, and what you would do in the first 3–6 months.
Evidence of skills: GitHub, a project portfolio, papers, Kaggle, or one strong reproducible project.
Two sentences on fit: which of our papers or models you have read, and what you would build on.

Step 2: apply to MBZUAI

MSc and PhD applicants apply through the MBZUAI application portal. The programs closest to our work are the PhD and MSc in Computational Biology.

After you submit, MBZUAI invites you to a mandatory online screening exam, and selected applicants are then invited to interview with faculty. Requirements, dates and fees are set by the university: always check the PhD admissions and master’s admissions pages before you apply.

Postdoctoral researchers, visiting students and remote interns do not use the graduate portal. Please see Research Position.

How I evaluate applications

Fit: does your interest match what the lab does?
Evidence of ability: strong projects, code, publications or rigorous coursework.
Clarity: can you explain a technical idea simply?
Research mindset: curiosity, persistence and willingness to iterate.

After screening, I may send a small technical task, such as reading a paper or implementing an idea, and schedule a short call.

12) Suggested reading

If you are new to PhD research culture, these are worth your time:

And to get to know our work: the MedMO and DoAtlas-2 pages, and the publication list.