Sew · Research
Accelerating Materials Science
with Machine Learning
Explore publicationsAtomistic simulation
Graph Neural Networks for
Machine Learning Interatomic Potentials
Graph neural networks can accelerate atomic-scale simulation. Machine learning interatomic potentials support structure relaxation and molecular dynamics of BaCd₂P₂, a promising photovoltaic semiconductor.
Using the MLIP to relax structures before DFT calculations reduced computation time in the reported study. The work also includes hybrid-functional DFT datasets assembled over several years.
Read the supporting research ↗The reported molecular dynamics simulation completed in under 7 hours. Results are specific to the BaCd₂P₂ study described in the April 2026 thesis abstract.
Simulation, synthesis & characterization
Discovering New Optoelectronic Materials
Combining first-principles calculations with optical experiments to understand promising semiconductors, their defects, and their potential in solar energy.
The research spans BaCd₂P₂, CaZn₂P₂, CaCd₂P₂, and ZnP₂.
Read the researchData for materials discovery
HSE06 Dataset
Hybrid-functional DFT data for machine learning and materials research. Explore sample materials, crystal structures, band gaps, and calculation files in the public preview.
The full dataset is available upon request.
Machine Learning Research & Software
Wef ML · C++ deep learning library
A custom tensor and neural-network library with CPU, Vulkan, and CUDA implementations. It includes attention, convolutional layers, sequential and functional model APIs, and examples for image classification and translation.
Explore the source code ↗Diffusion models & language models
Wef’s model-development projects include nano-DDPM for image generation and image or video super resolution, alongside AmharicGPT. The existing Wef demos showcase this work.
View Wef’s model demos ↗AI for organic synthesis planning
Machine learning supports reaction prediction, while graph search connects predictions into synthesis routes. This ACS chapter reviews these approaches, including methods based on natural language processing.
Read the 2023 chapter ↗Publications
Recent materials research and selected work in machine learning.
Radiative transitions in BaCd₂P₂ solar material
Gideon Kassa et al. · Journal of Applied Physics
Experimental and first-principles investigation of band-edge and defect-related light emission in a promising solar absorber.
Read paper ↗BaCd₂P₂: A Promising Impurity-Tolerant Counterpart of GaAs for Photovoltaics
Gideon Kassa et al. · Advanced Optical Materials
Optoelectronic measurements show the potential of BaCd₂P₂ prepared from relatively low-purity starting materials.
Read paper ↗Artificial Intelligence Based Organic Synthesis Planning for Material and Bio-Interface Discovery
Gideon Kassa et al. · ACS Symposium Series, Chapter 6
A review of machine learning for reaction prediction, natural language processing, and graph search for synthesis planning.
Read chapter ↗Publication status and source links reviewed October 17, 2026.
Wef’s supporters
