👋 Hi, I'm Ajay — a Postdoctoral Research Associate at Los Alamos National Laboratory (LANL) 🚀, working at the intersection of machine learning interatomic potentials (MLIPs), graph neural networks (GNNs), and nonadiabatic molecular dynamics.

My research focuses on building AI surrogates that make quantum chemistry simulations tractable at scale — replacing expensive ab initio calculations with uncertainty-aware ML models for ground and excited-state dynamics. What I work on:

• 🤖 MLIPs & Active Learning — HIP-NN / HIPPYNN architectures with uncertainty quantification for automated nonadiabatic AI-MD
• ⚡ Excited-State Dynamics — Trajectory surface hopping (FSSH), multi-state PES, phase tracking for nonadiabatic coherence
• 🔬 Hybrid QM/MM — Absorption and fluorescence spectroscopy in explicit environments; ensemble FC methods
• 💊 Drug Discovery — Protein-ligand binding free energy pipelines, BTK/KRAS-G12C inhibitor modeling (Frontier Medicines)🔬🎛️
• 📐 Ab Initio Materials Design — Excitonic coupling in π-stacked aggregates; energy transfer via Marcus, Förster, and Redfield theories

Stack: HIP-NN · GNN · PyTorch · NEXMD · CUDA · Ray · OpenMM · TeraChem · ORCA · AMBER · ASE · RDKit · SLURM

📩 Email | 💼 LinkedIn | 🌐 Portfolio | 🐦 Twitter | 📄 Google Scholar

📍 Los Alamos, NM · 📅 November 2024 – Present · Advisor: Prof. Sergei Tretiak

Building next-generation AI surrogates and machine learning interatomic potentials (MLIPs) to overcome computational bottlenecks in nonadiabatic molecular dynamics — enabling energy transfer simulations in large-scale molecular systems previously beyond the reach of quantum chemistry.

Key contributions:

• 🤖 Uncertainty-Aware MLIPs: Active learning pipelines built on HIP-NN / HIPPYNN / GNN architectures for automated ground, adiabatic, and nonadiabatic AI-MD with out-of-distribution detection
• ⚡ Surrogate Models for Multi-State Dynamics: ML surrogates for excited-state PES using trajectory surface hopping (FSSH) with rigorous phase tracking across S0–S3 states
• 🧪 Excitonic Coupling in PDI Aggregates: Quantitative structure-property relationships in perylene diimide (PDI) trimers — covalent tethering controls electronic coupling for organic electronics. Published in Nano Letters (Top 10%, 2026)
• 🌀 X-ray Circular Dichroism: Mechanistic framework for chiroptical signal design in azobenzene derivatives. JPCL Letters (Cover Article, Top 15%, 2025)
• 🎙️ MLCM-26: Co-organizing Machine Learning in Chemical & Materials Sciences 2026 — Santa Fe, NM · May 18–21 · $50K+ raised · 40 speakers · 100 participants

HIP-NN HIPPYNN PyTorch NEXMD CUDA Ray FSSH GNNs Active Learning SLURM

📍 Merced, CA · 📅 August 2018 – August 2024 · Advisor: Prof. Christine Isborn

Engineered high-fidelity QM/MM computational workflows for simulating absorption and fluorescence spectra of chromophores in explicit solvent — solving the long-standing challenge of capturing both vibronic coupling and environmental broadening simultaneously.

Key contributions:

• 📊 Ensemble Franck-Condon Spectroscopy: First extension of E-FC methods to fluorescence spectra; first direct comparison of all three E-FC variants across NBD, Nile Red, and 7MC in explicit QM/MM solvent. Demonstrated Eopt-avgFTFC achieves gold-standard accuracy at 25% of computational cost. Published in J. Chem. Phys. (2024)
• 💡 Polaritonic Chemistry ($7.5M DOD-funded): Computed excited-state energies, transition dipole moments, and Franck-Condon spectra for J- and H-type molecular aggregates in collaboration with experimental groups at UC San Diego and Penn State. Published in Nature Communications (2022)
• 🔬 Explicit Solvent QM/MM Pipelines: Full AIMD sampling using TeraChem + AMBER (GAFF2) with electrostatic embedding, CAM-B3LYP/6-31G(d) + TDA-TDDFT, 100 uncorrelated snapshots per trajectory

QM/MM AIMD TeraChem AMBER TDDFT Franck-Condon Gaussian OpenMM Python

📍 South San Francisco, CA · 📅 May 2023 – August 2023 · Manager: Monika Williams

Applied hybrid QM/MM methods to accelerate structure-based drug discovery pipelines for two high-value oncology targets at a leading targeted protein degradation biotech.

Key contributions:

• 🐍 Automated Pipeline: Python workflow for SMILES-to-desolvation energy calculations and conformational sampling for BTK inhibitors using OpenMM + TeraChem QM/MM interface
• ⚖️ Binding Free Energy: QM/MM free energy pipelines for accurate rank-ordering of BTK inhibitors; combined with MM-PBSA for reliable binding affinity predictions
• 🎯 KRAS-G12C Modeling: pKa and lipophilicity analyses for MedChem and synthetic chemistry teams — directly informing lead optimization for one of oncology's most sought-after targets
• 🔩 Molecular Docking: Biased and unbiased ligand-based docking using MOE across multiple BTK inhibitor scaffolds

OpenMM TeraChem MOE RDKit QM/MM Python Free Energy KRAS-G12C BTK

📍 Bangalore, India · 📅 January 2018 – July 2018 · Advisor: Prof. Biman Bagchi

Investigated Förster Resonance Energy Transfer (FRET) beyond the classical point-dipole approximation — targeting a fundamental limitation in how excitation energy transfer is modeled in biological imaging and solar energy systems.

Key contributions:

• 🔬 Beyond Förster Theory: Relaxed point-dipole and fixed orientation factor constraints for more realistic energy transfer modeling in donor-acceptor systems
• 🌈 Spectral Benchmarking: TDDFT excited-state geometry optimization and absorption/fluorescence spectra for AlexaFluor-488/594 (bioimaging) and Triphenylamine-Rhodamine (dye-sensitized solar cells) in DMSO and DCM
• 📐 DFT Functional Benchmarking: Demonstrated basis set effects were minimal for rigid dye systems; identified B3LYP as optimal functional achieving close agreement with experimental spectra

Gaussian TDDFT B3LYP DFT Python Implicit Solvent FRET Excited-State

📊 Google Scholar · 🔬 ResearchGate Score: 70.5 · 📄 ORCID: 0000-0002-8313-1393

0. 🔄 Building Uncertainty-Driven Machine Learned Interatomic Potentials for Automated Non-Adiabatic Molecular Dynamics

Ajay Khanna, ..., and Sergei Tretiak

Designing uncertainty-aware MLIPs with active learning pipelines built on GNN/MPNN architectures for automated ground, adiabatic, and nonadiabatic AI-MD — enabling reliable out-of-distribution detection for multi-state molecular dynamics simulations.

MLIPs GNNs Uncertainty Quantification Active Learning Nonadiabatic MD HIP-NN

1. Covalent Control of Excitonic Interactions in Perylene Diimide Trimers: A Computational Study

Ajay Khanna, Jean-Hubert Olivier, Sebastian Fernandez-Alberti, and Sergei Tretiak, Nano Letters, Top 10%, 2026

Established quantitative structure-property relationships in π-stacked PDI trimers (Free, Sandwich, Zigzag configurations), demonstrating how covalent tethering strategies control electronic coupling heterogeneity — providing a design blueprint for next-generation organic electronics and photovoltaics.

Excitonic Coupling PDI Aggregates π-Stacking Organic Electronics TDDFT Energy Transfer

2. Deconstructing Chirality: Probing Local and Non-local Effects in Azobenzene Derivatives with X-ray Circular Dichroism

Ajay Khanna, Victor M. Freixas, Lei Xu, Niri Govind, Jeremy R. Rouxel, Marco Garavelli, Shaul Mukamel, and Sergei Tretiak, Journal of Physical Chemistry Letters, Cover Article · Top 15%, 2025

Developed a site-specific X-ray spectroscopy framework to decouple local vs. non-local chiroptical contributions in azobenzene derivatives — establishing a mechanistic foundation for the rational design of functional molecular machines and chiral materials.

X-ray Circular Dichroism Chirality Azobenzene Molecular Machines Chiroptical NEXMD

3. Calculating Absorption and Fluorescence Spectra for Chromophores in Solution with Ensemble Franck-Condon Methods

Ajay Khanna, Sapana V. Shedge, Tim J. Zuehlsdorff, and Christine M. Isborn, Journal of Chemical Physics, 2024

First extension of ensemble Franck-Condon (E-FC) methods to fluorescence spectra; first direct comparison of all three E-FC variants across NBD, Nile Red, and 7MC in explicit QM/MM solvent. Key finding: Eopt-avgFTFC achieves gold-standard accuracy at 25% of the computational cost — recommended method for production spectral simulations.

QM/MM Franck-Condon Fluorescence Absorption Spectroscopy AIMD Vibronic Effects TeraChem

Degree	Institution	Focus
🔬 Ph.D., Computational Chemistry	University of California, Merced	Hybrid QM/MM · AIMD · Fluorescence Spectroscopy · Adiabatic MD · Resonance Energy Transfer
⚗️ M.Sc., Chemistry	National Institute of Technology, Rourkela	Computational Chemistry · Zintl Clusters · Advanced Experimental Chemistry
🔭 B.Sc. (Hons.), Chemistry	University of Delhi	Physical · Organic · Inorganic Chemistry
💻 Diploma, Web Design & Development	eZone Academy, New Delhi	Web Development · UI/UX Design

Certification	Issuing Organization	Year	Skills
🤖 Microsoft AI & ML Engineering	Coursera	2026	AI Engineering · ML Pipelines · Azure ML · Model Deployment
🧪 Introduction to Cheminformatics & Medicinal Chemistry	Udemy	2023	Cheminformatics · Drug Discovery · QSAR · RDKit
📊 Data Science with Python	Simplilearn	2022	Python · Data Analysis · Scikit-learn · Pandas
🚀 Fundamentals of Accelerated Computing with CUDA Python	NVIDIA	2022	CUDA · GPU Programming · Parallel Computing · Numba

A modular Python library for multi-trajectory NVE/NVT benchmarking of MLIPs against quantum chemistry references. Designed for systematic evaluation of machine learning interatomic potentials across ground and excited states.

Key features:

• Multi-trajectory NVE/NVT ensemble comparison
• Quantitative agreement metrics: KL divergence, Wasserstein distance
• Publication-quality plotting pipeline
• Modular architecture for easy extension to new MLIP frameworks

MLIPs HIP-NN HIPPYNN Benchmarking NVE/NVT Python NumPy Matplotlib

Co-organizing an international conference at the intersection of machine learning, chemistry, and materials science.

• 40 invited speakers from academia, national labs, and industry
• 100 participants · Intentionally small for deep scientific exchange
• Rising Star ⭐ talks for early-career researchers
• GPU compute awards + SCM Research Excellence Awards
• Registration deadline: April 17, 2026

MLCM26 LANL MLIPs GNNs Materials Science Computational Chemistry Conference

End-to-end pipeline for uncertainty-aware nonadiabatic molecular dynamics using HIP-NN/HIPPYNN architectures with active learning — enabling automated multi-state excited-state simulations at a fraction of the cost of ab initio methods.

Key features:

• Trajectory surface hopping (FSSH) with phase tracking across S0–S3
• Uncertainty quantification for out-of-distribution detection
• Ray-based hyperparameter optimization for HIPPYNN models
• NEXMD integration for photochemical systems

HIP-NN HIPPYNN FSSH NEXMD Ray CUDA Active Learning Uncertainty Quantification

Last updated: April 2026