Osium AI will boost the materials development cycle with Machine Learning

French startup Osium is looking into materials science research and development as a use case for machine learning. The startup has recently raised $2.6 million in a successful seed round with participation from Y Combinator, Singular, Kima Ventures, Collaborative Fund, Raise Phiture, and several angel investors, including Hugging Face CTO Julien Chaumond. Currently, Osium is composed of the co-founding duo Sarah Najmark and Luisa Bouneder.

Najmark and Bouneder have a strong background in Physics and Chemistry and plenty of AI expertise. Najmark told TechCrunch that while completing research in cosmetics, she realized that research and development was slowed down by methods relying on intuition and a lot of trial and error. After graduation, Najmark joined Google X (now X Development) and worked as Tech Lead and Project Manager. During her time there, she authored two machine learning patents. Similarly, Luisa Bouneder noticed how relying on intuition and trial and error hindered the materials development process while working on data products for industrial companies.

The duo started having conversations with contacts in the industry and academia, and they realized that intuition-based manual methods were not the only challenges holding materials development back, as there were also sustainability concerns. In a nutshell, materials science is not only about developing the newest, lightest, and most durable materials ever seen but also about using optimized and greener processes that save time and reduce the amount of waste produced by seemingly neverending rounds of trial-and-error development.

Osium AI's solutions are data-driven tools that optimize the back-and-forth between materials formulation and testing. The company website states that Osium's machine-learning models can assist R&D teams with accelerated microscopic image analysis, materials property prediction, and inverse design. In the case of property prediction, for instance, the model receives a list of criteria as input and then helps teams predict the resulting material's properties based on the information received. The models can also help developers refine and optimize new materials without needing extensive trial and error testing.

Najmark and Bouneder claim that after talking with their industry and academic contacts, some industrial companies are already trying out Osium's models and finding a 10x acceleration of materials analysis and development. As part of the co-founders' plan to turn these contacts into business clients, they plan to invest some of the raised money into growing their team. Osium is already accepting demo requests via its webpage. Osium may very well be getting started, but it is clear that the startup is in the perfect position to become a game-changer in the field.