Scandium Labs is an AI research and engineering company building machine learning systems for computational materials science.

Our mission is to accelerate the discovery and development of next-generation materials by combining advances in artificial intelligence, graph neural networks, scientific computing, and physics-informed machine learning.

We develop open research infrastructure, reproducible machine learning systems, and production-grade software for materials property prediction, scientific discovery, and foundation models for materials science.

Our work focuses on the intersection of artificial intelligence and computational materials science, including:

• Graph Neural Networks (GNNs)
• Physics-Informed Machine Learning
• Materials Property Prediction
• Solid-State Battery Materials
• Solid Electrolytes
• Scientific Machine Learning
• Representation Learning for Materials
• Foundation Models for Materials Science
• AI for Scientific Discovery
• Reproducible ML Infrastructure

We believe research software should be engineered with the same rigor as production software.

Our projects prioritize:

• Reproducibility
• Scientific correctness
• Performance
• Maintainability
• Experiment tracking
• Transparent evaluation
• Documentation
• Modular system design

Scandium Labs develops software and research infrastructure for:

• Materials property prediction
• Multi-task learning for scientific data
• Physics-informed neural networks
• Graph-based representation learning
• Large-scale materials datasets
• Experiment tracking and reproducibility
• High-performance training pipelines
• Model evaluation and scientific benchmarking

Our current research is centered on scalable machine learning systems for materials discovery.

This includes:

• Physics-informed graph neural networks
• Multi-task learning for material property prediction
• Automated data pipelines
• Large-scale materials datasets
• Experiment management infrastructure
• Training optimization
• Scientific benchmarking
• Model evaluation and analysis

Where appropriate, Scandium Labs develops and maintains open-source software that supports the broader scientific and machine learning communities.

Repositories are designed with an emphasis on:

• Clear architecture
• Comprehensive documentation
• Reproducibility
• Extensibility
• Research-grade implementations

Core technologies used across our projects include:

• Python
• PyTorch
• PyTorch Geometric
• NumPy
• SciPy
• Docker
• GitHub Actions
• CUDA
• Scientific Python ecosystem

Every project aims to satisfy the following principles:

• Reproducible experiments
• Versioned datasets
• Configurable training pipelines
• Automated experiment tracking
• Comprehensive evaluation
• Performance profiling
• Clean software architecture
• Publication-quality documentation

We welcome collaboration with researchers, engineers, and organizations working at the intersection of machine learning and materials science.

If your work aligns with our research interests, we encourage you to open an issue or start a discussion in the relevant repository.

Scandium Labs was founded by Shamique Khan.

GitHub: https://github.com/shamiquekhan

LinkedIn: https://www.linkedin.com/in/shamique-khan/

Scandium Labs

Artificial Intelligence for Materials Discovery.