MCP and AI tooling
MCP servers for AI assistants, content-plan generation via the Claude API, LLM-powered support analytics with Ollama, n8n.
I build web apps, APIs and the infrastructure behind them: backend, queues, databases, Kubernetes and CI/CD. On the web since 2010 — I started out with microcontrollers and Pascal.

Space, ↑ or tap the field to jump. Speed ramps up slowly.
MCP servers for AI assistants, content-plan generation via the Claude API, LLM-powered support analytics with Ollama, n8n.
Moved MySQL and MongoDB from Yandex Managed to our own DigitalOcean droplets: replication, TLS, daily backups to Spaces with restore scripts.
A shared GitLab CI template for every service: deploys to multiple clusters, builds with docker buildx bake, skips the build when the image already exists.
Prometheus and Grafana instead of Zabbix, self-hosted Sentry, monitoring alerts in Telegram.
A kanban dashboard that runs Claude Code and Cursor agents on tracker tasks, each in its own git worktree, from analysis to merge request. The plan gate is enforced: code starts only after a human approves the plan. A safety hook blocks push, kubectl and access to secrets. Plain Node, no framework, a single dependency.