Pioneering the future of artificial intelligence through cutting-edge language and diffusion models
We're developing advanced LLMs for niche use-cases where most frontier models fail. We have experience with RAGs, Graph RAGs, LORAs, finetunes, and even training our very own dLLM models where conventional solutions reach their limits. An important part of our LLM research are many experiments focused on reducing the lifetime cost of these models to make them economically viable even in small sectors.
We're consistently working on improving our expertise in diffusion models. Diffusion is a revolutionary technology, that enables massive performance benefits, better and implicit reasoning, and is even the most architecturally elegant solution in many cases. The most common use of it is in image-gen, but it's also used experimentally in other forms of generative ai. On top of it, we're experimenting with using it in more abstract state spaces.
An important part of our research is creating voice syntesis models as it's one of the hardest types of gen-ai to localize. We're focused on creating a scalable procedure for training localized variants of popular voice synthesis models with minimal overhead cost.
A diffusion-based language model project focused on preserving intellectual legacy and advancing Czech-language AI research.
AI-powered semantic search through Czech court decisions. Find relevant case law by keywords, case descriptions, or even legal documents. Currently covering Ústavní soud with more courts coming soon.
At ClosedBrain, we're committed to advancing the frontiers of artificial intelligence through rigorous research and innovation.
Our multidisciplinary team of researchers and engineers work collaboratively to develop AI systems that are not only powerful but also aligned with human values and needs.