How we work
Work at Orobix is the result of a close interplay of three teams.
Our mathematicians distill knowledge from data by leveraging on the latest deep learning advancements.
Our engineers break down complex problems and physical phenomena and take our solutions into the real-world.
Orobix is first and foremost an engineering company. We're here to create enabling technology that allows humans address hard problems. To achieve this, we cross-breed artificial intelligence with engineering, bridging top-down modeling approaches with bottom-up data-driven reasoning.
Deep learning is one of our main focuses. While we use it across the board, we specialize in challenging image recognition and segmentation tasks using fully convolutional networks, augmented with semi-supervised capabilities, so that we can limit the amount of annotated data required by training deep neural networks to be compatible with typical real-world settings.
Effective communication of results to end-users is key to the adoption of advanced technologies. We leverage on dimensionality reduction techniques and advanced visualization to create embeddings that help explain autonomous decisions to humans, or build similarity spaces to put information into context.
On the engineering side, we have a proven track record on modeling physical phenomena using partial differential equations and building effective processing pipelines to turn multidimensional data into computational domains and boundary conditions. We use robust computational geometry algorithms and advanced visualization techniques to have non-expert users access complex, quantitative information and make informed decisions.
As a company that creates end-to-end solutions, we develop a lot of software in-house. We are polyglots, with a marked interest in functional programming. In our daily work we span from lower-level languages, such as C/C++, when performance is key, to expressive functional languages like Clojure to build clean, scalable systems. We build our frontends on portable HTML5/WebGL technologies and leverage on Torch and TensorFlow for deep learning.