What we do

• Machine learning, NLP – we design and train models (neural networks or classic approaches, e.g., based on Logistic Model Tree)) that can analyze data, draw conclusions, and support decision-making processes. We feel most comfortable in the area of natural language analysis. We create domain-specific language models largely based on transformer architecture, but we also have experience in training models similar to ChatGPT. We always adapt the technique used to solve a problem to the problem itself.
• Data analysis, dataset preparation – one of our main specialties is the structuring of unstructured data. We create dedicated knowledge bases and training datasets, which we can support with a large-scale augmentation process carried out in our laboratory. The process is based on our proprietary solution, which we are constantly developing in new directions.
• Software engineering – naturally, all solutions result in a functioning system. We mainly limit ourselves to backend work, which is why our work often results in a complete solution architecture (e.g., Docker containers) with an exposed access API (Rest) for the system being created.

We approach each problem individually. Regardless of what you read in the literature, with us you will probably do it differently :). We always work closely with the client, who is an active participant in the project. Often, due to the risk of an individual approach, the client becomes an expert whose task is to evaluate partial results. This allows us to control and limit the propagation of errors to subsequent stages of development. It should be added that, in our understanding, software development is divided into three main stages, which are presented in the graphic. We carry out the entire process in our own server architecture, away from cloud solutions. From model training and testing, the CI/CD process in the development environment, to acceptance testing in a dedicated test environment. For production, we use our own Docker image repositories with controlled access per user. We ensure security at every stage and provide support in maintaining our solutions after implementation.