Labify

Inspiration

It started in the lab. While testing a new lattice material Betool designed, she realized we had no available data about its mechanical properties: no elastic modulus, no yield strength, no stress-strain curves. Characterizing it needed expensive equipment, access to specialized testing labs, long print and setup times, and hours of repetitive data analysis. Everything was slow, manual, and error-prone.

When we finally managed to manually test it, the results were SHOCKING. The material outperformed conventional foam structures in energy absorption. Yet, we almost missed that discovery because of the barriers in the traditional testing pipeline. That moment defined the problem:

Science is moving faster than our ability to test, understand, and validate what we create.

The Problem

Today, AI platforms like the Berkeley Materials Project Lab and Google DeepMind’s GNoME have predicted millions of potential materials: over 2.2 million structures and nearly 400,000 stable entries. But most of these exist only as digital entries. Without accessible simulation, validation, and analysis tools, their potential remains locked away.

The problem is that raw chemical data alone doesn’t create impact. Without tools to simulate, analyze, and interpret the data’s impact, we’re sitting on a treasure trove of discoveries we can’t yet use.

That’s where Labify comes in.

What it does

Labify bridges the gap between AI prediction and physical validation. It combines The Materials Project database and insights from DeepMind’s GNoME model with automated FEA (Finite Element Analysis) and CFD (Computational Fluid Dynamics) simulations, allowing users to visualize, test, and analyze materials virtually before they ever reach the lab.

Through an interactive 3D interface, researchers can explore lattice geometries, run performance simulations, and compare results in real time. The system automates repetitive testing workflows and provides AI-driven insights about material suitability for real-world applications from aerospace to biomedical design. It translates complex material science language into natural language that’s easy to understand and work with. This gold mine of stored data will allow scientists to generate insights so that it learns from the behaviours of materials and better improves the quality of data available for usage. This is for scientists, engineers and anyone looking for guidance on how to create their next frontier!

How we built it

Challenges we ran into

• Translating incomplete crystal data into simulation-ready geometries.
• Handling large data and computational complexity while maintaining speed and accessibility.
• Building a 3D visualization interface that is intuitive yet powerful.
• Balancing scientific accuracy with usability for non-experts

Accomplishments that we're proud of

What we learned

We learned that translating material lingo into natural language is a huge gap in the industry and there’s so much more it to than meets the eyes.

More importantly, we learned that the next frontier in materials science isn’t about generating more data it’s about making sense of what we already have. The future belongs to labs that can learn autonomously, test virtually, and extract meaning faster than ever before.

Labify is how we get there.

What's next for Labify

While we tried to make our simulation as precise as possible (eg: crystal lattice generation), there is always more room to make it more precise.
To improve the predictions by the model, it will be helpful to find ground-truth data regarding mechanical properties and perform RL post-training of our LLM to improve the predictions it make.
In the future, we would also want to automate the hardware for the physical lab experiments to predict a material’s properties