Clinical Data Scientist
2019 - Present (10 months)
IKNL is the quality institute for oncological and palliative research and practice. IKNL collaborates with healthcare professionals, managers and patients on the continuous...more improvement of oncological and palliative care.
Consultancy project at Zuyderland (Medical Center).
- Developed phone chargeback system for monthly allocation of phone bill consumption costs per user and respective...more cost centre. - On over 3k mobile contracts, reduced 10% of the total monthly subscription costs by eliminating/redefining contract terms according to observed consumption patterns.
Configured back office interface for data management and visualization.
- Improved data collection processes and processed existing sources for increased data utility....more - Configured back office interface and created data visualization methods for actionable insights.
Data Science intern
2017 (2 months)
Created a recommendation algorithm for connecting entrepreneurs with their ideal mentors to guide them through their innovative endeavours.
Developed an evolutionary algorithm which seeks to identify the optimal hyperparameter configuration for a given deep learning task.
Bravis Hospital Bergen op Zoom
2016 (4 months)
Developed a system to extract information from unstructured referrals to evaluate skin cancer diagnosis in primary care.
Industrial Engineer intern
2016 (3 months)
Analysed internal practices in a manufacturing company and provided guidance for optimising a wide range of operations, including business economics, information management,...more human performance, operations research and product development.
Data Scientist, Python · Working with data containing sensitive information requires one to carefully consider privacy enhancing methods, in order to… · More extract value safely without identifying any individuals. Over the last decade, academics have actively sought to find stronger definitions and methodologies to achieve data privacy while preserving high data utility. Furthermore, numerous recent privacy breaches and upcoming European data protection regulations (GDPR) urge organisations to strive for privacy enhancing practices. Often techniques proposed by academics can be considered complex and restrictive. For instance, one can restrict analysts from underlying data by only releasing aggregate statistics. Preferably, one would have to adapt their practices as little as possible and thus be able to work with data in its traditional format, while not risking any individual's private information. This data could then safely be shared, analysed or used for testing purposes. One way to achieve this premise is to generate private synthetic data, mimicking statistical patterns and characteristics of the raw dataset in a privacy-preserving manner.
Being able to move to London and work for a high potential startup on such an interesting research project has been very rewarding. In particular, the project seems to represent all facets of data science very well. Synthetic data generation remains a complex process and is only rarely applied in a real-world context. It requires a strong statistical background in order to understand how certain patterns in raw data can be captured. Moreover, implementation in academic papers is often limited and requires one to carefully reason what their limitations might be on real-world data sources. Finally, as these algorithms tend to be slow the challenge remains to find new ways of optimisation to make their approaches more scalable. All these elements combined made for a very satisfying learning experience.
Algorithm recommending mentor and entrepreneur partnership
Data Scientist, Python · Bridge for Billions is an online incubator with the objective of impulsing early-stage projects from around the world. One key… · More component of their offering is to provide entrepreneurs advice from industry experts. In this project, we created a decision support system which provides recommendations for strong potential mentor & entrepreneur partnerships.
During my studies, I felt I needed more experience applying the theory I had learned in a real professional environment. Therefore, I decided to do a second internship voluntary during my time in Madrid. I reached out to an ambitious startup working in a field which to me seemed to provide real value by giving people the tools to succeed in their innovative endeavours. They were very interested in attaining more value out of their data. In particular, they aimed to move beyond intuitive decision-making and use the rich data they have to acquire new insights and automate decision processes. As a result, we implemented a scalable recommendation algorithm, allowing them consider hundreds of potential partnerships, while determining the characteristics that would predict strong matches and in the end actually bring people together.
Hyperparameter optimisation through evolutionary algorithms
Data Scientist, Python, R Statistical Programming Language · Deep learning is an exciting field that allows computers to solve complex problems through… · More advanced networks in a self-learning manner. One major challenge during the deployment of these models is selecting the correct hyperparameter settings. For instance one has to carefully consider the architecture structure, as in number of neurons, layers, learning rate types, regularisation methods and training batch sizes. Finding the optimal parameters has been considered an art in itself.
Here an evolutionary algorithm can assist this selection process by automatically generating multiple generations consisting of various parameter settings, Individuals in these generations evolve over time, resulting in a final generation containing the optimal hyperparameter selection for a given deep learning task without the need for any human interference. Thus achieving higher accuracy and limiting the need for expert knowledge.
Alongside my Master studies I got to work on a deep learning optimisation project for a Consultancy by Telefónica in Madrid. I had the pleasure to work and learn from very inspiring people in a Spanish professional environment. In a short period of time, I had to learn about various deep learning neural networks and optimisation methods. Finally after doing an extensive literary review, we decided to implement the Differential Evolution algorithm, achieving over 20% increase in accuracy compared to previously employed methods.
An automated system for diagnosis information extraction
Research Scientist, Python, R Statistical Programming Language · A system enabling healthcare institutions to automatically extract relevant diagnosis… · More information from unstructured referral letters. Subsequently, the extracted information is compared to the diagnosis listed in specialist reports. Hence, one can evaluate the diagnosis history of general practitioners automatically to ultimately asses their referral practices.
As a kid, I was determined to avoid hospitals at all cost. Working in the medical field was thus completely unimaginable. Hence, you can imagine my surprise to end up doing my first major research project at the dermatological department of the Bravis hospital. Nevertheless, this experience has been incredibly valuable, as it allowed me to work with people from a completely different profession but who had a similar feeling that their rich datasets, if analysed and modelled properly, could provide immense benefits to advance healthcare and limit the burden of mundane repetitive tasks.
Here I also encountered my love for Data science and the ability to derive new insights through discovering hidden patterns. I got to learn programming on my own, in two different languages, in a very intense and short period of time while also becoming acquainted with the field of information extraction and medical data analytics.
In the end we ended up with a system which was able to process historic information records, extract relevant information, integrate it with other sources and ultimately evaluate the quality of the contents inside. It allowed us to asses the current state of referral practices by examining early GP diagnoses within unstructured referral letters and comparing them to final diagnosis listed in specialist documents. As a result, we were able to perform normally long and repetitive research studies in a fraction of the time and cost.
Optimising processes, systems and organisational structures
Industrial Engineer · An extensive report detailing opportunities to optimise operational processes, information flows, organisational structures and… · More innovative practices within a manufacturing company.
Acted as a bridge between all disciplines within a submersible pump manufacturing company, allowing me to understand the problems each person faces during their daily work. These interactions, combined with my technical background, allowed me to analyse and write an extensive report detailing process, system and organisational alterations that would optimise internal practices while increasing human performance and satisfaction. I particularly enjoyed working with people of diverse backgrounds ranging from engineers to managers, and accountants to sales staff, which ultimately gave me solid understanding of their work. One starts to truly appreciate the complexity that comes with producing such a highly innovative product.
What I Do
Data Scientist and Industrial Engineer, approaching the end of his double degree Master in Data Science. People-oriented, strong public speaker and attracted to innovation in collaborative teams. I strive to take a multidisciplinary perspective while tackling complex data-oriented problems.
During my bachelor thesis research, I worked on a research project at a Healthcare institution regarding clinical data processing issues. Not only was the medical environment completely unknown to me, I also had very little experience regarding data analytics, programming and natural language processing. In a short and intense period of four months I stepped into the field of Healthcare analytics and demonstrated my ability to rapidly learn two programming languages (i.e. Python and R) to ultimately develop an information extraction system operating on unstructured data, while writing an extensive Bachelor Thesis. This effort was awarded with a excellent grade of 9 out of 10. All these aspects combined made for a demanding, but very fulfilling research experience.
A next step in my career as a Data Scientist from August 2018 onward. Prepared to move anywhere and learn new skills. Eager to work in a place which stimulates continues learning, collaboration, and encourages product ownership.
Playing, writing, recording and regularly performing live with a band.
Which founders or startups do you most admire?
My father. He always demonstrates strong technical knowledge, work ethic, negotiation skills and great sacrifice to run his company successfully, while still being there for us whenever we need him.
What's your super power?
Being able to relate to a variety of perspectives, allowing me to step beyond my strong mathematical, statistical and technical background to connect with a very diverse range of people.
What's something you wish you had done years earlier?
Having training judo intensely in my youth, I was very happy to rekindle my lost passion by coming into contact with Brazilian Jiu-Jitsu in Sweden. BJJ has fascinated me ever since, encouraging continuous learning, humility and strategic thinking.