Not demos or pet projects — real platforms in production. What it was, what we built, what it delivered.
Problem. The client needed a platform that connects third-party GPU machines, hands them to customers for workloads, and takes over all accounting and billing.
Solution. A system of 10 services: node registration, job orchestration, usage metering, automatic billing and a control panel. All containerized, with monitoring and self-recovery.
Problem. A stretch goal: the site had to survive complete loss of its server with zero human involvement.
Solution. Monitoring detects the outage → a script rents a new VPS → cloud-init provisions the stack → a neural network fills the site with content → traffic switches over. The whole cycle takes about a minute.
Problem. Dozens of physical devices across locations — issues had to surface before customers noticed.
Solution. A single health-check system with metrics and logs, Telegram alerts with severity levels. The on-call sees the problem in chat and reacts by runbook — often before any complaint.
Tell us what hurts in your infrastructure — we'll propose a solution and estimate within a day.