Machine learning for chemistry
Teaching OLMo-7B chemistry
Continued pre-training on ChEMBL SMILES, evaluated across eight MoleculeNet tasks.
AI in science
AI in science, from code to evidence
I explore deep learning, machine learning, and scientific computing through chemistry models, generative systems, and astronomical data. Each project connects the method to inspectable code and results.
Explore the researchThree ways into the work
Deep generative models
MolGAN in PyTorch
A from-scratch guide to graph generation, WGAN training, and reward optimization.
Scientific computing
JWST spectra and light curves
Working with telescope data to inspect an exoplanet transit and atmospheric spectrum.
Latest notes
Compare molecular graphs and SMILES for chemistry AI through MolGAN and OLMo-7B, including how each representation changes the model pipeline.
Continued pre-training OLMo-7B on chemical SMILES, followed by MoleculeNet classification and regression tests with explicit evidence limits.
A documented Streamlit prototype for summarising legal documents and asking follow-up questions with Gemini.
MolGAN PyTorch implementation guide covering molecular graph generation, the R-GCN discriminator, reward network, WGAN training, and QM9 setup.