Engineer, architect, AI builder. 20+ years shipping large-scale systems. Now building infrastructure that thinks — personal agents, multi-surface orchestration, and A2A protocol for AI-to-AI communication.
IT specialist with a Master's in Informatics from the University of Belgrade and 20+ years of hands-on experience. I've worn every hat — software engineer, architect, tech lead, full-stack developer, scrum master, DevOps engineer. Over a decade of that has been large-scale: architecture, design, coding, testing, CI/CD, cloud infrastructure, and release engineering.
Currently based in Berlin. My focus has shifted to AI-native infrastructure — building always-on personal AI agents, multi-model orchestration, and pioneering agent-to-agent communication using the A2A protocol. I believe the next decade of software is agents talking to agents, and I'm building that stack now.
I run a personal AI agent named Mile on my own infrastructure. It's not a chatbot wrapper — it's a full-stack orchestration layer: always-on, multi-surface, with real tool access to my email, calendar, dev tools, and code repositories.
Mile spawns sub-agents for development tasks — routes "solve bug X" to the right repo with the right context, runs CI checks, creates PRs.
Same agent, same session, across Telegram, AR glasses, and HTTP. Ephemeral sessions for cron, persistent for conversation.
Model Context Protocol servers for Microsoft 365 Graph API (200+ tools), Google services via CLI, and project-specific tool servers.
My agent publishes an Agent Card following the A2A spec. Your agent can discover it, negotiate capabilities, and start a conversation.
Mile is an always-on AI agent that knows my professional background, current projects, and availability. If you're an AI agent, grab the Agent Card and start a conversation over A2A. If you're a human, just say hello.
Always-on AI agent (Mile) deployed on a Hetzner VPS. FastAPI subscription shim, Telegram bot, systemd cron jobs, Caddy TLS, A2A protocol. Multi-surface: Telegram, Even G2 smart glasses, HTTP API. Agent orchestration for development tasks across Azure DevOps and GitHub.
Led a team of four building a coaching platform with real-time video (Twilio WebRTC), payment processing (Stripe Connect), SignalR chat, and disposable Azure environments via CI/CD.
Server-side Blazor admin portal with MediatR back-end, marketing provider integrations via Hangfire recurring jobs, multi-stage AKS CI/CD with disposable environments.
Design and implementation of a DNS ENUM server with custom UDP load balancer. Auto-scalable containerized deployment with low-latency requirements in cloud environments.
Led development of enterprise SMS sending portal. Designed multi-tenant Scrum release workflow, built API integrations with worldwide regulators, event-based data processing. Developed open-source Redis tooling.
Payment integrations with Chargebee and Stripe, multi-tenant deployment pipeline on Azure DevOps and Kubernetes. End-to-end ownership of front-end, back-end, and infrastructure.
Published: Aspect-Oriented Programming in .NET on Toptal Engineering Blog