T-Technologies on Open Source in AI/ML: Inside the LLM Development Process

T-Technologies has openly shared how internal AI/ML work is organized within one of Russia's largest technology structures. The interview was given by two key leaders of the direction: Anatoly Potapov, who heads the group of fundamental LLM technologies, and Daniil Gavrilov, who leads research in artificial intelligence. T-Technologies is the technological core of the T-Bank ecosystem.

In recent years, the company has been actively building up AI/ML competencies: from applied models for banking products to fundamental research in large language models. The team's task is not simply to implement ready-made solutions, but to develop its own technologies and participate in shaping the open source ecosystem around them. Anatoly Potapov oversees the direction that works with the basic components of the LLM stack: architectures, training methods, infrastructure for working with large models.

According to him, a focus on open source allows not only to accelerate their own development, but also to attract strong engineers — those for whom it matters that their work is visible and reproducible beyond the company. Daniil Gavrilov is responsible for the research part: experiments, publications, search for new approaches. He describes the team structure as a hybrid of an academic laboratory and a product department — with an emphasis on ensuring that research has practical output rather than remaining only on paper.

Both leaders agree on one thing: openness is not a PR move, but a working tool. When internal developments are published in open access, this creates external feedback, helps find errors earlier, and builds a reputation that converts to hiring. This format — when a large Russian company publicly discusses the internal kitchen of AI development — is still rare.

The interview with the T-Technologies team is interesting not only as an industry case study, but also as a signal: Russian corporate AI sector is developing its own culture of openness, albeit cautiously.