Tubhyam Karthikeyan

spectroscopy deep learning open source cheminformatics

View Work

Get in Touch

What I Research

Vibrational Spectroscopy

Molecular Rep. Learning

Reaction Engineering

Information-Theoretic Chemistry

Physics-Informed ML

ML Infrastructure

drag to rotate

In Progress 2026

Information-Theoretic Limits of Spectroscopic Molecular Identification

A group-theoretic and information-theoretic framework for bounding what vibrational spectra can reveal about molecular structure. Establishes fundamental limits on spectroscopic identifiability using representation theory and rate-distortion theory.

information theorygroup theoryidentifiability

In Progress 2026

Hybrid State-Space Attention for Multi-Task Vibrational Spectroscopy

Hybrid state-space and attention architecture combining wavelet-domain embeddings, selective state-space sequence modeling, Mixture of Experts routing, and Variational Information Bottleneck for multi-task prediction across IR, Raman, and mass spectrometry.

state space modelsspectroscopymixture of experts

Planned 2026

Beyond Ground State: Benchmarking Equivariant GNNs for Excited-State Property Prediction

Systematic evaluation of eight equivariant graph neural network architectures on excited-state molecular properties including singlet/triplet energies, oscillator strengths, and fluorescence wavelengths, with transfer learning and scaling analysis.

graph neural networksexcited statesbenchmarking

Planned 2026

InvyrA: Conditional Molecular Graph Generation from Vibrational Signatures

A conditional generative framework that inverts the spectroscopy problem — given an input IR or Raman spectrum, generates candidate molecular graphs using Spekron latent embeddings and an equivariant graph diffusion decoder, with calibrated uncertainty over the combinatorial solution space.

inverse problemsmolecular generationdiffusion models

IR Spectrum — Benzaldehyde

research — preprocessing.py

theorem — symmetry quotient

VIB Head

Variational Information Bottleneck — disentangle chemistry from instrument artifacts

Ibuildopen-sourcetoolsforcomputationalchemistry—fromspectralanalysislibrariestoneuralnetworkarchitecturestointeractivesimulations.

What I Build

SpectraKit

Published on PyPI

01/03

▌

PythonPyPIMIT

View project →GitHub &nearr;

Latest Writing

From the Blog

Feb 2026

Can You Hear the Shape of a Molecule?

Feb 2026

Hello World — Building My Corner of the Internet

Read all posts →

project — file structure

training — spectral_fm.py

diff — predict.py

before

after

Exploreinteractivesimulationsthatbringmolecularphysicstolifeinyourbrowser.

Interactive Simulation

VibeScope

Real-time3Dvisualizationofhowatomsmove—exploremolecularvibrationsinteractively.

Select a molecule, pick a vibrational mode, and watch atoms oscillate in real time. Built with Three.js and WebGL for real-time rendering.

Launch VibeScope

vibescope — simulation.ts

WebGL Three.js Real-Time

molecule — structure viewer

BenzeneC₆H₆

Normal Mode Explorer

Beginner

Decompose molecular vibrations into individual normal modes. Animate each mode in real time.

Try it →

Predicted Spectrum

Spectrum-to-Structure

Advanced

Input a vibrational spectrum and watch as an ML model predicts the molecular structure in 3D.

Orbital Architect

Game

Build atoms by filling orbitals. Aufbau, Pauli, and Hund's rule are the game mechanics.

Play now →

Symmetry Explorer

Educational

Visualize point groups, symmetry operations, and character tables. See how symmetry determines selection rules.

Try it →

Featured Three.js + WebGL

VibeScope — Full 3D Molecular Dynamics

The flagship simulation. Select any molecule, pick a vibrational mode, and watch atoms oscillate in real time. IR/Raman spectrum overlay, adjustable amplitude, and GIF export.

WebGL Three.js 60fps

Launch VibeScope →

pipeline — spectral_fm

IngestQM9S + ChEMBL

PreprocessWavelet + Norm

Train4x RTX 5090

EvaluateR², RMSEP

DeployPyPI + API

Tools & Languages

Python

PyTorch

CUDA

RDKit

Docker

AWS / GCP

Mamba / SSMs

Graph Neural Nets

drag to rotate

Target R²↑ good

0.000

QM9S Dataset↑ good

GPU Hours→

0.000

Papers WIP↑ good

ci/cd — project health

PackageStatusVersionCoverageTests

coding patterns — generative

113 active cells — procedurally generated

Mon

Wed

Fri

Less

SPECTROSCOPY • DEEP LEARNING • OPEN SOURCE • CHEMINFORMATICS • GROUP THEORY — SPECTROSCOPY • DEEP LEARNING • OPEN SOURCE • CHEMINFORMATICS • GROUP THEORY — SPECTROSCOPY • DEEP LEARNING • OPEN SOURCE • CHEMINFORMATICS • GROUP THEORY — SPECTROSCOPY • DEEP LEARNING • OPEN SOURCE • CHEMINFORMATICS • GROUP THEORY — SPECTROSCOPY • DEEP LEARNING • OPEN SOURCE • CHEMINFORMATICS • GROUP THEORY — SPECTROSCOPY • DEEP LEARNING • OPEN SOURCE • CHEMINFORMATICS • GROUP THEORY —

Let'sbuildsomethingtogether.

Send a message

availability — status

┌─────────────────────────────┐
│ Tubhyam Karthikeyan │
│ ICT Mumbai '30 │
└─────────────────────────────┘

Google Scholarktubhyam

HuggingFacektubhyam

All Projects Read the Blog