ML Engineer · Python Developer
I'm focused on Machine Learning, Data Science and building practical ML-based systems.
Mostly working with:
- recommender systems
- computer vision
- tabular ML
- feature engineering
- ranking models
- ML pipelines
- backend services around ML models
I like projects where ML is not just a notebook, but a working pipeline: data preprocessing, validation, model training, inference logic and integration with an API or product.
- AIIJC / AI Challenge — prize-winner, School Track
- National Technology Olympiad, Artificial Intelligence track — prize-winner
ML / DS:
Python, pandas, NumPy, scikit-learn, CatBoost, PyTorch, TensorFlow, SciPy
RecSys / Ranking:
CatBoostRanker, ALS, BM25, item-to-item similarity, learning-to-rank, reranking
Computer Vision:
OpenCV, ViT, image preprocessing, color correction, histogram-based models
Backend / Tools:
FastAPI, Flask, Django, Docker, PostgreSQL, SQLite, MongoDB, Git
Languages:
Python, SQL, C++
Computer Vision solution for the Automatic White Balance problem in mobile cameras.
The task was to predict a distribution of possible white points in a scene under different lighting conditions.
What I used:
- Vision Transformer backbone
- image metadata
- log-chroma histogram features
- edge maps
- depth features
- Gaussian Mixture Model parameterization
- Wasserstein/KL-based loss
Tech: Python, PyTorch, OpenCV, NumPy, timm, image processing
Result: 3rd place, AIIJC School Track
Recommender system solution for the NTO AI Final 2025/26.
The task was to recover lost positive user-book interactions after a logging failure.
The solution is based on a two-stage RecSys pipeline:
- candidate generation
- feature engineering
- CatBoostRanker
- PU-learning logic
- time-window validation
- final top-20 ranking
Candidate generators included:
- ALS
- BM25
- popularity baseline
- item-to-item similarity
- metadata-based retrieval
- incident-window generators
Tech: Python, pandas, implicit, CatBoost, scikit-learn
Score: ~0.144289 NDCG@20
Personalized book recommendation system for the AI Academy hackathon “По страницам”.
The task was to generate top-20 book recommendations for each user while balancing relevance and genre diversity.
The final pipeline included:
- handcrafted user/item features
- text and graph features
- collaborative filtering features
- CatBoost / LightGBM / XGBoost models
- neural network meta-features
- CatBoostRanker with YetiRank
- diversity-aware MMR reranking
Tech: Python, pandas, CatBoost, LightGBM, XGBoost, PyTorch, implicit, sentence-transformers
Score: 0.719
Telegram bot for automating educational course workflows.
The bot handles:
- course registration
- user interaction through inline keyboards
- certificate generation
- contact information
- Google Sheets integration
- PostgreSQL storage
Tech: Python, aiogram, PostgreSQL, Google Sheets API
- Telegram: @main4562
- GitHub: github.com/AndHunter
- Kaggle: andrewsokolovsky