About

Alchemist Labs is where I write about building software that still has to work after the demo: distributed systems, production infrastructure, AI, and the engineering judgment that connects them.

I'm Stefan Mojsilović, a Munich-based systems engineer. My work sits at the boundary between application systems and the infrastructure beneath them: distributed services, high-performance storage and networking, bare-metal and cluster orchestration, GPU/HPC workloads, and the tooling used to operate all of it reliably.

I build mostly with Elixir/OTP, Go, Rust, Python, and the occasional amount of shell that reality requires.

I care about the unglamorous details: failure domains, data paths, observability, deployment safety, capacity, and what happens at 3 a.m. when a system meets a condition its design document did not anticipate.

Lately, I have been especially interested in AI systems that leave the demo environment: agents with real tools, production model deployments, and the guardrails, evaluation, and operational feedback loops needed to make them genuinely useful. I am less interested in whether a system can produce a convincing answer than whether it can be trusted with meaningful work.

The recurring themes here are the BEAM, distributed systems, AI in production, and technology choices that age gracefully. Expect concrete examples, strong opinions held provisionally, and an honest account of the trade-offs.

This site is also an experiment: a Phoenix application built with LiveView and Hologram, with posts authored in Markdown.

If you're building reliable AI systems or infrastructure with real operational consequences, you'll probably find something useful here.