Curriculum Vitae
Ashkan Pirmani. I make evidence out of health data that cannot be gathered in one place. I also build the systems that let other groups do the same. Six years across registries, ordinary care records and consortia in more than thirty countries. From September 2026 I do this inside a company. The roles closest to it sit between applied machine learning, real world evidence, and the engineering underneath both.
Download CV (PDF)02 September 2026, 294 KB01Impact
The four results worth knowing
Each one links to the paper it came from, and says what kind of result it is. Nothing here is a projection unless it says so.
7.2% and 31% better
Fitting a federated model to each country beat the same method without that fitting, at ranking who will get worse and at catching them. It also beat training on everything pooled together, by 3.8%. Measured in simulation on records from 26,246 patients, 283,115 clinical episodes, contributed by 146 centers. I conceived the study, ran the experiments, analyzed the data and led the writing. First author, in a collaboration of 73. Funded by the Flanders AI Research Program.
45.5% more records to work with
Designed and built the federated route into a study that already had two other ways in. Five registries joined through it that could not have joined otherwise. The four counted in the paper added 3,527 records to the 7,757 already collected. Deployed and run by the registries themselves. This is a data feasibility result first. The question could not have been answered on the records that were already reachable. The collection was run by the MS International Federation and the MS Data Alliance. Its findings went into the global covid advice for people with multiple sclerosis. First author on the architecture paper.
Half the sites, almost all the score
On a public benchmark, letting half the centers stay centralized instead of forcing all four to federate kept 97.5% of the fully federated score. That is the degree of federation, and it is the idea I am known for. A working prototype, evaluated on Azure by us.
A public resource, open and maintained
FLkit is live: 39 pages, eight sections, 11 ways in depending on your role, seven worked project stories, 29 contributors. Supported by ELIXIR Belgium. These are numbers about what is in it, not a measurement of how many teams use it. That evaluation has not been done yet.
02Experience
What I have done
Innovation Mandate, with Johnson & Johnson Innovative Medicine
An Innovation Mandate from Flanders Innovation and Entrepreneurship, the Flemish government's innovation agency. The grant goes to a consortium of one company, one university and one researcher. Two years. The first develops the method at the university. The second applies it with the company, which co-funds it and sets the direction, and includes working inside it. Treatment comparison from routine records, where no trial exists.
- Starts September 2026. Nothing here has happened yet.
Postdoctoral Researcher
KU Leuven and Hasselt University, Belgium.
- Lead FLkit. It went from an internal need to a public resource other groups can use without asking me.
- Designed and built FLoRank, the method behind the ECTRIMS grant. Taking it from a prototype to its first run on a real data network.
- Single point of contact across more than thirty countries, between the people who hold the data, the people who treat the patients, and the people who decide what is allowed.
- Supervise master's students and teach the data science course.
Doctoral Researcher
KU Leuven and Hasselt University, Belgium.
- Built the federated route into the worldwide multiple sclerosis and covid data sharing effort, so registries that could not send data could still take part. It became the largest group of its kind.
- Author of FL4E, and of the degree of federation.
- Ran the first systematic comparison of personalized federated learning for two year disability progression. Published in npj Digital Medicine, 2025.
Research Assistant, Decision Support Systems Lab
IranDoc, Iranian Research Institute for Information Science and Technology, Tehran.
- Agent based and system dynamics simulation of the national parcel network, searched with a genetic algorithm for a better set of hubs. Best scenario: 15% lower projected transport cost. Never built, so never realized.
- Delivered as a decision support dashboard and a policy analysis to top level executives.
03Systems
Selected systems
What each one achieved is in section 01. This is what state it is in and where to find it. Not all of it is a repository.
Lead developer. Public and maintained.
FL4E and FL4E-Analysis
Author. Reference implementation and the benchmarks behind the paper.
Author. The study code for the npj paper.
Global Data Sharing Initiative, federated route
Architect. Not a repository. Infrastructure that ran inside five registries that were not allowed to send data, so they could take part anyway.
04Education
Where I trained
Dual PhD
Electrical engineering, KU Leuven. Biomedical sciences, Hasselt University. Belgium. Defended November 2024.
- Thesis: From Centralized to Federated. The Journey of Data in Healthcare.
- Supervisors: Yves Moreau (KU Leuven), Liesbet M. Peeters (Hasselt University), Niels Hellings (co-supervisor).
MSc, Industrial Engineering
Socio-economic systems. Kharazmi University, Tehran.
MBA, Quality Engineering
Tose'e Institute, Tehran.
05Awards
Recognition
Innovation Mandate, Flanders Innovation and Entrepreneurship
Awarded to one researcher, with a company and a university behind them. The second year is not automatic; it depends on passing a review of the first. The work it funds is under Experience.
ECTRIMS personal grant for scientific quality
Awarded for FLoRank, presented at ECTRIMS 2025, Barcelona.
Fully funded PhD scholarship
Both doctorates were funded in full, KU Leuven and Hasselt University, through to the defense in 2024.
06Teaching
Teaching and supervision
Data Science in Healthcare
Teaching member, Healthcare Engineering program, Hasselt University. Academic years 2022 to 2023 and 2023 to 2024.
Master's thesis supervision
Seven Master's theses in Artificial Intelligence, KU Leuven.
- Predicting admission to care. Federated, pooled and local data compared.
- How splitting the data changes what the model learns.
- Federated boosting in semi-supervised learning.
- Contrastive learning for federated models.
- Personalized models. Architectures and parameter exchange.
- Fine-tuning a global federated model for local performance.
- Early disease detection from shallow whole genome sequencing of cell-free DNA.
07Areas
Areas I work in
Every paper, abstract and chapter is on the publications page. Talks and news are on their own page.
Methods
- Federated learning and federated analysis
- Personalization under non-identical data across sites
- Privacy-preserving machine learning
- Survival analysis and classification on clinical data
Data
- Real world and observational data
- Cohort building and phenotype definition across registries
- Registry harmonization across countries, and data quality
- Data feasibility: whether a question can be answered at all
- OMOP common data model, through OHDSI
- FAIR data practice and standardization
- Low prevalence and rare disease research