Revolutionize and pioneer advanced materials innovation

Inspiration

The traditional approach to materials discovery is slow, expensive, and heavily reliant on trial-and-error experimentation. The growing demand for novel materials in aerospace, clean energy, electronics, and healthcare has inspired us to build AI for Science models. Our goal is to enable high-throughput materials design and process optimization, accelerating the discovery of next-generation materials that meet real-world industrial demands.

What it does

DeepQuantumX combines quantum mechanical simulations and generative AI models to discover optimal and stable materials unforeseen by human, design synthesis pathways, and optimize processes to improve manufacturing. Tailored for researchers and industrial R&D teams, it accelerates the materials discovery process and enhances production scalability.

How we built it

We fine-tune state-of-the-art diffusion-based generative models with our proprietary datasets to explore unique and new materials. We also develop machine learning potentials to enable high-throughput sampling on large-scale systems with quantum mechanical accuracy. The backend is powered by Python and PyTorch, with APIs to interface with open-source databases, and the frontend is built for intuitive interaction with material simulations and design workflows.

Challenges we ran into

Aligning quantum simulation data formats across multiple databases. Training robust generative models that respect physical constraints. Building an interface that’s intuitive for both academic researchers and industry users. Ensuring scalability for large material spaces while maintaining accuracy.

What we learned

The synergy between domain knowledge and machine learning is crucial. Pure black-box models fall short when it comes to scientific complexity, necessitating the embedding of physical principles. High-quality training data is still one of the biggest bottlenecks in materials ML. Many of our competitors tend to rely on proprietary data rather than crowdsourcing. Researchers prefer tools that seamlessly integrate into their existing workflows. Most are risk-averse and product-driven.

What's next?

Scaling up generative AI models for broader material systems and classes. Partnering with industry to co-develop proprietary models for solving niche problems. Deploying a cloud-based platform for global access to our AI-powered design tools.

Built With

• hpc
• python