Hi, I'm Andrij Vasylenko.
I develop machine-learning methods and computational workflows that help scientists explore complex systems and make better decisions — faster. My work connects physics-based simulation, generative models, and experimental data into closed-loop pipelines for materials discovery. My work is supported by national HPC allocations and international collaborative grants, with partners in the UK, Switzerland, and the US.
I develop generative models and active-learning frameworks that guide experimental discovery of new materials. A variational autoencoder trained on crystallographic data learns compositional and structural patterns across chemical space, identifying unexplored elemental combinations as synthesis targets. Coupled with Bayesian optimisation and first-principles simulation, this approach has reduced experimental screening iterations by over 50% and supported discoveries reported in Science (2024) and Nature Communications (2021), with a UK patent pending.
Selected Publications:
A. Vasylenko et. al, Digital Discovery 4, 477 (2025)
G. Han, A. Vasylenko et. al, Science 383, 6684 (2024)
A. Vasylenko et. al, The Journal of Chemical Physics 160, 5 (2024)
A. Vasylenko et. al, npj Computational Materials 9, 164 (2023)
A. Vasylenko et. al, Nature Communications 12, 5561 (2021)
View projects
First-principles simulation remains the ground truth for materials properties — but it is expensive. I develop machine-learning models trained on quantum-mechanical data to reach DFT-level accuracy at a fraction of the compute cost, and use physics-based constraints to improve the interpretability and out-of-distribution reliability of generative models. This combination of simulation and ML underpins the closed-loop discovery workflows that have produced experimentally validated results across thermodynamic stability, electronic structure, and transport phenomena.
Selected Publications:
A. Pshyk, A. Vasylenko et. al, Communications Materials 7 (2026)
G. Han, A. Vasylenko et. al, JACS 143, 18216 (2021)
A. Vasylenko et. al, ACS Nano 12, 6023 (2018)
A. Vasylenko et. al, Phys. Rev B 95, 121408(R) (2017)
I contribute to open-source scientific software used well beyond materials research. This includes algorithmic contributions to PyOD — a Python library for anomaly and outlier detection with over 16 million downloads — and research pipelines for experimental data capture, ML workflow orchestration, and reproducible HPC-scale computation. See projects.
I also maintain the Materials Innovation Factory group repository
(50+ projects).