Software Engineering graduate (BSUIR, Minsk) transitioning into ML/DS.
Building production-ready machine learning services with deployment, testing, and CI/CD.
Passionate about applying ML to real business problems — from credit risk to predictive maintenance and LLM-powered systems.
ML & Data Science
Python scikit-learn CatBoost XGBoost LightGBM PyTorch pandas numpy MLflow LangChain NLTK
Statistics & Analysis
A/B testing hypothesis testing feature engineering EDA matplotlib seaborn
Deploy & Backend
FastAPI Docker docker-compose GitHub Actions uvicorn REST API
Languages & Tools
Python SQL C# C++ Git pytest Jupyter Linux
Production-style Applied AI MVP for automating customer support in a fictional electronics store.
The system combines a Telegram bot, FastAPI backend, PostgreSQL, n8n automation workflows, local RAG, intent classification, guardrails, SLA tracking, human-in-the-loop review, and a Streamlit analytics dashboard.
The project is designed as a realistic internal automation prototype rather than a simple OpenAI wrapper: critical decisions are handled by deterministic rules, database lookups, confidence thresholds, retrieval grounding, and escalation policies. OpenAI support is optional; the MVP runs locally with Docker Compose and mock/local mode.
Production-like ML monitoring service for credit scoring: FastAPI, PostgreSQL, Evidently drift reports, prediction logging, alerts, baseline management, model performance tracking, retraining triggers, Docker, CI/CD.
Production-style Computer Vision project for industrial safety monitoring: YOLO detection, PPE compliance checks, worker tracking, danger-zone alerts, FastAPI inference, Docker, training/evaluation pipeline, benchmarking, tests and CI.
Production-style NLP system for contract document analysis.
It classifies contract clauses, detects heuristic risk indicators, retrieves semantically similar clauses, and generates structured risk reports through a FastAPI service.
Built with a real LEDGAR data pipeline, TF-IDF + Logistic Regression baseline, optional transformer training pipeline, evaluation reports, semantic search, Docker, tests and CI.
The project connects my legal background with practical ML engineering and focuses on reproducibility, honest limitations, and human-in-the-loop document review.
Credit default prediction service on real banking data · 150k records
- Model: GradientBoosting · ROC-AUC 0.868 · PR-AUC 0.400
- MLflow experiment tracking · optimal threshold selection by F1 (+60% vs default)
- FastAPI + web UI + Docker + CI/CD GitHub Actions
Multi-agent market trend analysis system powered by LLM
- 5 specialized agents: Router · Researcher · Analyst · Extractor · Editor
- LangChain + FastAPI + Docker + async pytest + CI/CD
InsightData Analyst — a local RAG-powered SQL agent that translates natural language into SQL queries. Built with LangGraph, Ollama, and Qdrant: the agent retrieves real table schemas from a vector store, generates accurate SQL, and self-corrects on errors — all without sending data to the cloud.
Chest X-Ray Pneumonia Classifier - Detection of pneumonia in chest X-ray images using ResNet18 transfer learning.
📍 Minsk, Belarus
💼 Open to: remote DS/ML/DA positions · Russian companies · Relocation EU
