I’m Saurav Kumar Roy. I use this site to document hands-on work where machine learning meets chemistry, scientific computing, and open-source software.

Start with OLMO learns chemistry →

Chemistry and machine learning

OLMO learns chemistry

I continued pre-training OLMo-7B on the ChEMBL SMILES training split, then tested the result on MoleculeNet classification and regression tasks while documenting the limits of cross-paper comparisons.

Read the OLMO project notes

MolGAN

I implemented MolGAN from scratch in PyTorch, combining molecular graph generation with a reinforcement-learning objective for properties such as solubility and synthesizability.

See the MolGAN implementation

Scientific computing and open source

HEASOFT and X-ray astronomy

My scientific-computing notes cover setting up HEASOFT, configuring the NICER calibration database, and working with light curves, spectra, and FITS data.

Read the HEASOFT setup notes · Explore the JWST project

GSoC and OpenAstronomy

I also document my GSoC proposal work with Stingray and OpenAstronomy, including an interactive database concept for X-ray binary observations.

Read the GSoC proposal · Follow my open-source journey

Other experiments

The rest of the site follows the same build-and-document approach, from a latent-diffusion poem generator to an AI-assisted legal document analyser.

You can also find the underlying projects on GitHub or read my CV.