Code & data
I release the code and data behind my papers so that the results can be reproduced and reused. All repositories are on GitHub.
1. Machine-Learning Potentials for Thermal Conductivity
Workflows that use universal and fine-tuned machine-learning interatomic potentials to compute phonons and lattice thermal conductivity, with bonding analysis of the results.
- uMLIP_thermal_conductivity — fine-tuned MatterSim and other universal potentials for lattice thermal conductivity, with bonding analysis. · Paper: J. Mater. Chem. A 2026
- CHGNet_Thermal_condutivity — CHGNet with fine-tuning for lattice thermal conductivity; installable with
pip install chgnet_thermal_conductivity. - Li-Cu-Ti-Chalcotitanates- — machine-learning potentials for rattling and the bonding hierarchy in Li–Cu–Ti chalcotitanates. · Paper: J. Mater. Chem. A 2024
2. Explainable Machine Learning
Models that use chemical-bonding descriptors to predict a property and explain it.
- SCALP-Thermal-Conductivity — bonding-descriptor and conventional-feature models for the thermal conductivity of pnictogen chalcogenides. · Paper: Chem. Mater. 2025
3. Generative AI and Inverse Design
Code and data for generating new crystals and designing dopants for a target property.
- Generative_Materials — graph neural networks, cross-property transfer learning and CrystaLLM generation for low-thermal-conductivity materials. · Paper: ACS Appl. Mater. Interfaces 2026
- dopant_TE — dataset for machine-learning-driven dopant design of thermoelectrics. · Paper: ACS Appl. Electron. Mater. 2024
4. Literature Mining and Datasets
Pipelines that extract measured properties from published papers, with the source of every value recorded.
- AgSbTe2_data_extraction — pipeline and dataset from 173 papers on AgSbTe2 thermoelectrics (Seebeck coefficient, conductivity, κ, zT, dopants) with DOI provenance, plus a viewer app.
Software I work with
My workflows are built on open-source tools: pymatgen, MatGL and the MatPES dataset, ASE, Phonopy and phono3py, ShengBTE, TDEP and LOBSTER.