Viktor Masiuk

I’m a Data Scientist focused on turning machine learning into reliable, production-ready systems. I build end-to-end pipelines—from feature engineering and model training to containerized deployment, health checks, and monitoring—so models perform consistently outside the notebook.

Most recently, at Vester AI, I built a production-grade customer churn prediction system with a ZenML orchestration pipeline and artifact versioning. I trained and tuned an ensemble of sklearn models (XGBoost, LightGBM, RandomForest) with Optuna, achieving ROC-AUC 0.87, then exposed results via a FastAPI REST service (Pydantic schemas, health-check, async inference) and a Streamlit dashboard with SHAP-based explanations.

Alongside freelance-style projects and competitive hackathons, I’ve worked on applied ML and LLM-driven tooling—like a lost-pet identification system using OpenAI CLIP embeddings with ChromaDB and ZenML tracking. I also developed a Linux incident-response triage tool that extracts artifacts with Volatility 3, scores findings with custom sklearn models, and uses an LLM reasoning layer to generate structured IR narratives.