Translation. Region: Russian Federal
Source: Peoples'Friendship University of Russia
An important disclaimer is at the bottom of this article.
At the end of 2024, a research team of students from the Faculty of Physics, Mathematics, and Natural Sciences at RUDN University—Artur Busardykov, Mikhail Geller, and Kamil Mekhdiev—received a 4 million ruble grant from the Foundation for Assistance to Small Innovative Enterprises (FASI) to develop their startup, Skopeo.AI. Over the past few months, the team has not only refined the product but also gone from an idea to a ready-made solution, which is already being tested by its first partners.
Study, career, startup
The startup's founders combine their studies at RUDN University with work at large companies and their own project. Artur Busardykov, the team's driving force and mastermind, previously held the position of senior DevOps engineer at Innotech (a VTB project), but is now focused on Skopeo.AI and a couple of other personal IT projects. Mikhail Geller, who previously served as a lead DevOps engineer at ET Consulting (RosAtom), is also currently focusing on Skopeo.AI. Kamil Mekhdiev, having gained experience as a business analyst at BEORG, a company specializing in the development of intelligent systems based on neural networks, has moved to the position of team lead at VTB and continues to remain on the team.
"Experience with large IT organizations gave us an understanding of real business pain points: the cost of downtime, the labor-intensive nature of manual scaling, the complexity of multi-cloud management, and observability/audit requirements. This directly informed the Skopeo.AI architecture," says Artur Busardykov, Bachelor of Science (BSc) in the Faculty of Physics, Mathematics, and Humanities, majoring in Applied Mathematics and Computer Science.
From concept to working prototype
Skopeo.AI was conceived as a multi-cloud platform for managing Kubernetes clusters, helping businesses reduce cloud resource costs, improve infrastructure resiliency, and automate processes, eliminating the need for manual configuration.
Key features of Skopeo.AI:
Automatic scaling of cloud resources; load optimization and prevention of application downtime; monitoring and data analysis using artificial intelligence.
"The platform makes business infrastructure transparent, cost-effective, and efficient," says Artur Busardykov.
Today, the team can summarize its initial results. According to Artur, a working platform prototype has been built (cluster agents, server analytics, and a web panel). The MVP is in the final stages of refinement: the team is completing production scenarios for autoscaling and recommendations and polishing the UI. A pilot demonstration of the service is planned for the end of the year.
What was implemented with the 4 million ruble FSI grant?
The funds made it possible to create a fully functional platform framework:
Kubernetes agent (metrics collection, events, fault-tolerant delivery); server side: time series storage, load forecasting (Prophet/XGBoost/CatBoost), recommendation module; web panel: dashboards, forecasts with confidence intervals, list of recommendations and their application history, action audit; REST API and webhooks for integrations, basic economic analytics (assessment of potential savings); CI/CD, Helm charts, on-prem and cloud builds, security (TLS, RBAC, SSO/OAuth2).
Overcoming challenges
The months of development weren't without challenges. The team encountered several issues and found solutions. To eliminate noisy metrics and jagged time series, they added cleansing/normalization, an aggregation window, and confidence intervals, improving model robustness.
The team solved the problem of integrating into heterogeneous clusters by moving the setting to Helm-values and adding an autodetection component (metrics-server/kube-state-metrics/Prometheus). To balance automation versus control, they introduced policies with trust levels: "warn," "suggest," and "automatically execute with rollback."
The team has already conducted internal platform testing at partner facilities. The results showed that the recommendations effectively reduce CPU/RAM overhead. The goal is now to achieve a 30–50% savings from the service in real-world environments.
Team and partnership development
The project team remains compact but effective. It still consists of a Senior ML Engineer and two DevOps engineers. The startuppers also selectively engage external experts for project tasks (UI/UX, security) and developers.
An important achievement for the guys was the establishment of business connections.
"We're actively building and expanding our partner pool: we've already agreed on pilot projects with several companies and systems integrators, and we're continuing to seek new entry points and industry contacts. We don't plan to attract investors yet; we're focusing on expert support. One of our external advisors is experienced engineer and entrepreneur Mikhail Teplov: he helps us with mentoring and industry contacts," says Artur Busardykov.
The team's plans for the coming years are concrete and ambitious:
2025: Complete MVP, conduct 3-5 pilots, validate savings, release commercial pricing plans (on-premise SaaS), close key integrations (Prometheus/Grafana, GitOps, billing); 2026: Scaling sales (e-commerce, financial sector), expanding recommendation logic (SLA/budgets), federating multi-clusters, auto-remediation with secure "railguards," and cloud partnerships. The team also continues to participate in grant programs.
The students are currently preparing an application for the next stage of the competition, which is being held by the Federal Social Research Fund, and are looking at industrial tracks (IT accelerators, regional support measures).
The example of Artur, Mikhail, and Kamil is a story about how one can successfully combine studies, career, and entrepreneurship, creating innovations that can change the approach to working with technology.
Please note: This information is raw content obtained directly from the source. It represents an accurate account of the source's assertions and does not necessarily reflect the position of MIL-OSI or its clients.
.