I’m Yurii Havrylko, a Senior Machine Learning Engineer in Prague. I work on ML and AI systems that have to survive production: LLM and NLP pipelines, retrieval and RAG, and making inference fast and cheap — the gap between “the model works” and “the model works at scale without burning the budget.”

I build in the open: nerbench, an evaluation harness for zero-shot NER, and performance contributions to Presidio — benchmarked across five hardware platforms in this post.

This blog is where I write up benchmarks and engineering findings I wish someone else had published first.

Find me on GitHub · email.