Argyrios Kerezis.
Biomedical AI · Healthcare ML · Molecular bioinformatics

Selected project portfolio for research and health-tech roles.

I work with reproducible machine-learning workflows for medical imaging, clinical prediction, molecular bioinformatics, uncertainty estimation, and scientific computing. This page summarizes the projects most relevant to biomedical AI, pharmaceutical bioinformatics, and PhD-oriented research roles.

View selected projects GitHub LinkedIn

Relevant competence

Medical imaging ML3D segmentation workflows, MONAI transforms, validation metrics, visualization of predictions.
Clinical predictionStructured healthcare data, risk stratification, readmission prediction, length-of-stay modelling.
Molecular bioinformaticsQSAR-style Ki modelling, SMILES curation, RDKit descriptors, and chemically cautious validation.
Uncertainty & evaluationMC dropout, entropy/variance maps, calibration, limitations, clinically cautious interpretation.
Research codingClear pipelines, command-line scripts, reproducible setup, documented assumptions and outputs.

Selected projects

Medical imaging · Deep learning

Medical Segmentation Uncertainty

3D U-Net segmentation pipeline for the Medical Segmentation Decathlon hippocampus task, with uncertainty estimation using Monte Carlo dropout.

Repository →
ProblemImprove reliability assessment in limited-data medical segmentation.
ImplementationMONAI preprocessing, 3D U-Net training, sliding-window inference.
EvaluationDice, HD95, sensitivity, specificity, entropy, variance, calibration analysis.
Molecular AI · Cheminformatics

Ki Prediction from Molecular Structure

Code-first QSAR workflow for modelling molecular Ki values from SMILES using RDKit standardisation, descriptor engineering, and cross-validated scikit-learn models.

Repository →
ProblemModel binding-affinity signal from chemical structure in an Applied Pharmaceutical Bioinformatics setting.
ImplementationCanonical SMILES curation, largest-fragment selection, ECFP4, MACCS, RDKit, and custom descriptors.
EvaluationRegression on log10(Ki), activity-threshold classification, descriptor ablation, and explicit validation limits.
Clinical ML · Thesis

Long-COVID Patient Prioritisation

MSc thesis project on machine-learning-based prioritisation of Long-COVID patients for specialist consultation.

Repository →
ProblemSupport triage decisions under limited clinical capacity.
ImplementationSupervised ML, SVMs, class-imbalance handling with SMOTE.
RelevanceShows healthcare-data modelling with realistic clinical constraints.
Healthcare prediction · Prototype

Diabetes Readmission & Length of Stay

Applied clinical ML workflow for predicting 30-day readmission risk and length of stay in diabetic patients.

Repository →
ProblemEstimate readmission risk and expected length of stay from structured clinical data.
ImplementationFeature engineering, MLP/multitask modelling, FastAPI demo backend, Streamlit prototype.
LimitationsResearch and portfolio prototype; not a clinical deployment system.

Background

Education
KTH Royal Institute of TechnologyEngineering Physics / biomedical and data-driven research direction.
Education
National Technical University of AthensTranslational Engineering in Health & Medicine, with thesis work in clinical machine learning.
Research
AstraZeneca Molecular AIExperience connected to molecular AI and biomedical machine-learning work.

Technical skills

Core tools and methods used across the selected projects.

PythonPyTorchMONAIscikit-learnRDKitQSARpandasNumPyFastAPIStreamlitC++GEANT4Medical imagingClinical predictionMolecular bioinformaticsUncertainty estimationScientific computing

Languages

Greek — nativeEnglish — professional working proficiencySwedish — SFI kurs D

Contact

I am interested in research engineering, biomedical AI, healthcare machine learning, medical imaging, clinical prediction modelling, molecular bioinformatics, and PhD-oriented research roles. References and full CV available on request.

LinkedIn GitHub