Сергей Дадаев

I’m an ML Engineer with over 3 years of experience in Classic ML and NLP, focused on taking models from research to production-ready services. I specialize in developing end-to-end ML pipelines, with an emphasis on model optimization and practical, measurable impact.

In my current role, I own full-cycle development—from feature engineering and algorithm comparison (CatBoost, LightGBM, XGBoost) through hyperparameter optimization and time-series validation. I build Airflow DAGs for automated model retraining pipelines, wrap models into FastAPI services, and containerize with Docker for production deployment. I also implement business and technical metric monitoring using Prometheus and Grafana.

I’m especially proud of the production outcomes I’ve driven: I developed a PD model with ROC-AUC 0.81, increasing consumer loan approvals by 4% at a constant risk level. I deployed an income prediction model with MAPE = 35%, and reduced Average First Response Time by 40% using instant topic detection and response template selection. I’ve also automated client feedback and technical support processing via LLM-based sentiment analysis and intelligent tagging systems.

Previously, at Sakhalin Energy, I designed and deployed ML models in production using industrial sensor data, building automated data collection and processing pipelines. I developed a RAG system for semantic analysis and search across regulatory databases and training materials, and delivered results like reducing production downtime by 20% through predictive maintenance for control valve failures.