From October 6 to 11, an offline camp for the technical community AI HOUSE Camp 2026 was held in the Carpathians. For six days, 28 Machine Learning engineers, united in 7 teams, developed AI solutions for two critically important practical tasks: recognizing Russian military chevrons and anonymizing personal data in Ukrainian texts.
From October 6 to 11, an offline camp for the technical community AI HOUSE Camp 2026 was held in the Carpathians. For six days, 28 Machine Learning engineers, united in 7 teams, developed AI solutions for two critically important practical tasks: recognizing Russian military chevrons and anonymizing personal data in Ukrainian texts.
The event was partnered by SQUAD, a company specializing in the development of AI solutions for cameras. The participants were advised by leading experts from MacPaw, EPAM, Kyivstar, airSlate and other IT companies. In addition to intensive development, IT workers joined in charity work, raising over 70,000 hryvnias for the benefit of the Reaktivna Poshta NGO.
The first challenge was prepared in collaboration with the War Archive, which is creating an analytical system for automatic search and analysis of photo and video materials. The goal of the development is to automatically identify Russian chevrons in millions of frames from open sources, determine the units of the occupiers, and link them to specific war crimes.
To train the models, the participants processed over 15,000 images and a catalog of 462 reference chevrons. In this area, the jury recognized two teams:
Creators won for the highest quality solution. The team collected additional photos from Telegram channels, expanded the library of references, and built a map of connections between departments. Their development is now ready to be used for initial file markup.
Starlight — awarded for the most promising approach, which consists in decomposing the chevron into heraldic elements (shield shape, symbols, colors). This method allows you to easily scale the model and define new divisions without collecting thousands of additional photos.
The created solutions will be integrated into the War Archive system to process a multi-million data set, which will significantly facilitate the work of investigators and prosecutors.
The second challenge was to create a tool for automatic search and replacement of personal data (names, addresses, etc.) in Ukrainian texts, including court decisions. This is necessary so that government agencies and businesses can securely transfer documents to external AI models and cloud services without violating the law on personal data protection.
The winner of this task was the NoName team. The participants independently collected an additional sample of documents and created a transparent pseudonymizer architecture with clear indicators for different data types.
The AI HOUSE organizers and community mentors plan to organize a meeting of the winning teams with representatives of the War Archive and the Ministry of Digital Affairs to discuss further technical integration of the created solutions into state and analytical projects.
"We deliberately chose a challenge where the result is measured not by points, but by real impact. In six days, the teams did what often takes months in the industry," emphasized AI HOUSE CEO Vitaly Rozman.
Currently, AI HOUSE has already begun preparations for the next camp and is looking for partners among technology companies.
