AtomGPTLab

Welcome!

The AtomGPTLab is part of the Departments of Materials Science & Engineering (MSE) and Electrical & Computer Engineering (ECE), as well as the Data Science & AI Research Institute (DSAI) at Johns Hopkins University. The lab’s goal is to develop agentic AI frameworks for materials science applications, integrating multiscale simulation and experimental capabilities.

 

 

Research

Materials are everywhere!

The total number of possible materials is estimated to be around 10100, and the Processing-Structure-Property-Performance (PSPP) paradigm across different length and time scales makes materials design one of the most challenging scientific problems.

We use a combination of quantum mechanics, classical mechanics, machine learning, and experimental techniques to accelerate the design of target materials with desired properties.

Our focus is on functional materials such as high-temperature superconductors, semiconductors, solar cells, dielectrics, and piezoelectrics; in their perfect, defect, and interface forms.

We apply physics-based methods such as density functional theory, tight-binding, molecular dynamics, and finite-element modeling to develop large-scale databases, which then serve as precursors for machine learning training.

Since computers and machines do not inherently understand materials science, we develop sophisticated algorithms using both conventional and deep-learning methods, including fingerprinting, graph neural networks, and generative pre-trained transformer based approaches to create forward and inverse materials design frameworks that can accelerate discovery by 100 to 1000 times.

For physics-based, experimental, and machine-learning based methods, we also develop community benchmarks to measure and guide project success. Our research involves multi-modal data, including scalars (e.g., energetics, bandgaps), spectra (e.g., X-ray diffraction, EELS), images (e.g., STEM and STM microscopy), and text (e.g., peer-reviewed articles, documentation).

To democratize our research outputs, we frequently share:

    1. Code on our GitHub page
    2. Data on Figshare
    3. Benchmarks on our Leaderboard
    4. User-friendly chatbot on the AtomGPT.org website

Teaching

I will be teaching a course on AI for Materials starting in Spring 2026 (EN.510.463 & EN.510.663).

This course introduces core concepts and modern techniques in artificial intelligence and data science, with an emphasis on applications in materials science. Topics include classification, regression, clustering, and generative AI, using models such as random forests, neural networks (NN), convolutional neural networks (CNN), graph neural networks (GNN), and transformer-based architectures (e.g., GPT). Students will work with real-world materials datasets derived from multiscale modeling and experimental measurements, spanning tabular, image-based, and structural formats. Hands-on coding exercises and a final project will reinforce both conceptual understanding and practical applications.

Team Members

Principal Investigator:  Kamal Choudhary

Prospective students:

If you would like to join the research group, it is recommended to

    1. Email your CV/Resume to [email protected] with subject of email such as “Postdoc/PhD/Undergrad research application”
    2. Fill up this Google form
    3. Go through these YouTube videos
    4. Execute these Google Colab Notebooks to get familiar with the basics.

The videos and notebooks are intended to reduce repetition and ensure all students receive the same foundational information.

Publications

    1. The JARVIS infrastructure is all you need for materials design
    2. ChatGPT Material Explorer: Design and Implementation of a Custom GPT Assistant for Materials Science Applications
    3. MicroscopyGPT: Generating Atomic-Structure Captions from Microscopy Images of 2D Materials with Vision-Language Transformers

  1.  

  1.  

Linkedin Twitter Youtube