Computational Scientist · Argonne National Laboratory

AI, HPC, and Quantum Chemistry

Quantum mechanics provides a powerful foundation for predicting molecular and material behavior. Yet applying it to problems such as materials design, drug discovery, and reaction chemistry remains challenging because accurate calculations are computationally expensive and chemical spaces are enormous.

My research focuses on closing the gap between what our physical models can predict in principle and what we can calculate in practice. I combine artificial intelligence, high-performance computing, and quantum chemistry to tackle challenging problems of the 21st century.

Portrait of Murat Keçeli

AI + HPC + QC = Accelerated Scientific Discovery

Publications

Artificial Intelligence

Machine-learned potentials, agentic AI, retrieval-augmented generation, and AI evaluation.

Papers Code

High-Performance Computing

Scalable eigensolvers, automated workflows, exascale-ready software, and distributed training.

Papers Code

Quantum Chemistry

Accurate thermochemistry, coupled-cluster methods, anharmonic vibrations, and density functional theory.

Papers Code
Recent publications

Papers from the last two years

All papers
Open source

Software

argonne-lcf/ChemGraph

Agentic framework connecting natural-language requests to molecular construction, simulation (ASE, MACE, RDKit, and external QC codes), analysis, and reporting through LangGraph tools, a CLI, a Streamlit UI, and an MCP server.

Python ★ 156
PACChem/QTC

Quantum Thermochemistry Calculator: integrates Open Babel with MOPAC/NWChem/Gaussian/Molpro calculations and MESS partition functions to generate NASA polynomials automatically.

Python ★ 7
Autonomous-Scientific-Agents/IQC

Python framework for running quantum-chemistry workflows (single-point energies, geometry optimization, frequencies, thermochemistry) at scale with QC codes and machine-learned potentials (MACE, FAIRChem UMA), writing Parquet datasets for downstream analysis..

Python ★ 6
petsc/petscagent-bench

Agentified evaluation framework, built on A2A and MCP, for benchmarking code-generation agents that write PETSc programs..

Python ★ 6
All software
News

Recent updates

Notes

Writing

All notes