An agentic AI framework for computational chemistry workflows.
ChemGraph is an agentic framework that lets researchers plan, reason about, and execute atomistic simulation workflows through natural language. It combines graph neural network–based foundation models for fast, accurate calculations with large language models for task planning and scientific reasoning, wrapping simulation codes (DFT, Monte Carlo, machine-learning interatomic potentials), databases, HPC schedulers, and workflow engines behind an intuitive interface (Pham et al., 2026; Pham et al., 2026).
A core finding of the work is that decomposing complex tasks into smaller subtasks through a multi-agent design enables even smaller LLMs to match or exceed a single-agent GPT-4o baseline across a suite of computational-chemistry benchmarks.
ChemGraph is an agentic framework powered by AI and state-of-the-art simulation tools to streamline and automate computational chemistry and materials science workflows, combining graph neural network foundation models with large language models for planning and scientific reasoning.
@article{pham_chemgraph_2026,title={ChemGraph as an agentic framework for computational chemistry workflows},author={Pham, Thang D. and Tanikanti, Aditya and Ke{\c{c}}eli, Murat},journal={Communications Chemistry},year={2026},publisher={Nature Publishing Group},doi={10.1038/s42004-025-01776-9},}
Multi-Agent Orchestration for High-Throughput Materials Screening on a Leadership-Class System
Thang Duc Pham, Harikrishna Tummalapalli, Fakhrul Hasan Bhuiyan, and 5 more authors
@misc{pham2026multiagent,title={Multi-Agent Orchestration for High-Throughput Materials Screening on a Leadership-Class System},author={Pham, Thang Duc and Tummalapalli, Harikrishna and Bhuiyan, Fakhrul Hasan and V{\'a}zquez Mayagoitia, {\'A}lvaro and Simpson, Christine and Balin, Riccardo and Vishwanath, Venkatram and Ke{\c{c}}eli, Murat},year={2026},archiveprefix={arXiv},primaryclass={cs.AI},}
