Vitaly Bogachev

Python, TypeScript, LLM evaluation, RAG, fine-tuning, MCP, CI/CD, test automation

How to evaluate my own tools without fooling myself: frozen code, labels committed first, then run.

pr-witness: running the base branch's tests against AI-written pull requests. ftgate: checking that a fine-tuned model is served the bytes it was trained on.

AI engineering roles and consulting: LLM products in production, evaluation, agent tooling. Happy to talk about testing AI-written code.