Tsimur Bardziyan

At FlyRank, I supported production ML initiatives and framed a content dataset spanning about 145K pages and 28 clients as a ranking problem. I used historical data to predict future page declines.

I built features from trailing 90-day windows and used client-grouped GroupKFold evaluation to guard against leakage. The grouped evaluation scored 0.557 AUC, compared with 0.653 for a naive random split, and I helped translate the analysis into a review workflow covering 85K+ pages.

In my RAG System — AI Knowledge Assistant project, I built an AI assistant that searches user-provided documents and cites its sources. I developed the workflow from document ingestion and semantic retrieval through reranking and response generation.

I’ve also built projects for real-time ML inference, AI-assisted task automation, and model compression. I’m an undergraduate Artificial Intelligence student at Poznań University of Technology, interested in applying data and AI to real-world investment and sustainability problems.