Laboratory Chislitel and Informzashchita have introduced a software-hardware complex aimed at helping large Russian companies transition from repeated AI pilot projects to reliable industrial operation of machine-learning models.
The solution was announced at the TNF-2026 forum. At its core is a high-performance ML cluster paired with the Russian container orchestration platform Shturval. The developers stress that the offering is not merely another expensive GPU rack requiring manual configuration for every new project.
Instead, the complex converts raw compute capacity into a single managed resource. An ML specialist submits a task, and the platform automatically allocates CPU and GPU resources, provisions the required environment, connects storage, launches training jobs, and scales workloads as needed.
Automation is achieved through Kubernetes and established MLOps components including Kubeflow, MLflow, and model-serving tools. The overall architecture is structured in four layers: hardware infrastructure, the Shturval platform, the MLOps stack, and applied AI services ranging from forecasting and analytics to corporate assistants.
Supporting services around the core include IAM/SSO, object storage, an image registry, CI/CD pipelines, monitoring, and audit capabilities. Customers receive a unified compute pool, project isolation, centralized access management, and full model lifecycle support from experimentation through production deployment and updates.
ML tools interact with infrastructure only as a service rather than directly accessing hardware. According to project documentation, the architecture has already undergone industrial operation at a large state customer.
