Welcome to Latent Chemistry

Thanks for subscribing. I'm Tubhyam Karthikeyan, and this newsletter is where I share original research and technical writing at the intersection of machine learning, spectroscopy, and computational chemistry.

What to expect

• Deep dives into novel architectures for scientific data (state-space models, transformers for spectra, graph neural networks for molecules)
• Research notes from building SpectralFM -- a foundation model for molecular spectroscopy
• Tutorials on topics like self-supervised pretraining, optimal transport losses, and numerical stability in deep learning
• Occasional musings on the craft of research, scientific computing, and building in public

Why "Latent Chemistry"?

Every spectrum encodes a molecule's identity in a latent space of vibrations, rotations, and electronic transitions. The goal of this work is to learn those representations -- to decode the chemistry that hides in plain sight within spectral data.

The name also captures what I love about this field: the hidden structure waiting to be found, the chemistry latent in the math.

If you have questions, ideas, or just want to say hello, reply directly to this email. I read everything.

See you in the next issue.

-- Tubhyam

tubhyam.dev | GitHub | LinkedIn