I'm Yehor Kaliberda, an AI systems architect based in Aarhus, Denmark, working remotely with teams anywhere. I build autonomous AI systems that do real engineering work — type annotation, security remediation, code review — and I don't consider the work done until it has passed a verification gate that doesn't care how confident the model sounded.
That standard shows up in what I ship. FixProve checks every import and API call an AI assistant writes against what's actually installed, deterministically, with zero LLM tokens in the check itself. Patchward proposes security fixes that have to survive three sequential gates — the original scanner finding clears, the edit stays inside its authorized lines, and the project's own test suite passes — or the fix is discarded, not retried with a more persuasive prompt; a Patchward finding shipped downstream in checkdmarc v5.17.3. Provenar reads a Python repository and delivers complete PEP 484 type annotations as a single reviewable pull request, gated on zero pyright errors before it goes out; eleven of those pull requests have been reviewed and merged by open-source maintainers across four projects who owed me nothing.
I write about the reasoning behind those gates, and about what autonomous agents get wrong, in Philosophy. Everything I claim publicly links to an artifact: a merged pull request, a released package, or a named test — see Work for the full, honestly labeled list of what's shipped versus still in build.
I ship engineering work through CallMed AI. For engagements or questions, see Contact.