I am a highly motivated machine learning engineer with a strong interest in mathematics, engineering, and product development. With extensive experience in the complete life cycle of machine learning. My passion is to explore both the research and engineering aspects of machine learning to create innovative and practical solutions.
| Download CV |
Work experience
Machine Learning Engineer at Avito
- Worked on improvement various ranking algorithms in search
- Builded ETL pipelines to making datasets for different ranking algorithms
- Optimised and written efficient code for offline experiments
Machine Learning Engineer at Unleashing.ai
- Headed the engineering development of 2 projects with a copilot/chatbot and job recommendations
- Optimisation and design of code in projects to achieve real-time operation, scaling and distributed computing
- System design and setting up timelines for projects related to legal tech and job recommendations
Machine Learning Engineer & Product Owner at Memerest
- Developed scalable & stable FastAPI Recommendation API that reduces latency by 3x and memory usage by 10x
- Built from the ground up an analytics system for AB experiments
- Researched and implemented state-of-the-art ML models with a latency of 600ms for online recommendations
- Managed entire product with an average daily spend time app of 30 minutes, a 1-day retention rate of 45%, and 3K MAU
- Managed a team responsible for the development of a Mobile App for iOS and Android, including QA and Design
- Developed monetization strategy for a mobile app (0.9% CTR), while maintaining high retention and engagement levels
Machine Learning Engineer at Ozon
- Improving search ranking algorithms via new features, new methodologies, and new losses which increase GMV by 1%
- Optimizing data pipeline algorithms to speed up feature delivery by 1.5x
- Making tools for ranking explanations using SHAP values for the business admin page
Research Intern at Skoltech
- Construction Deep Graph Embeddings using nontrivial losses
- Improved quality of recommender systems using Graph Neural Network (GNN)
- New state-of-the-art approach for intrinsic dimension problem: GeoMLE
Junior Data Scientist at DIGINETICA
- Developing recommender systems using statistics, matrix factorization, DSSM and heuristics approaches
Research Intern at IITP(RAS)
- Parameters estimation in MMSB model, its application in community detection
- Classification problem with bio-medical data
Education
Skolkovo Institute of Science and Technology
Master’s degree: Mathematics and Computer Science
Department of Data Science
GPA 5.0/5
Department of Data Science
GPA 5.0/5
Moscow Institute of Physics and Technology
Bachelor’s degree: Applied mathematics and physics
Department of Radio Engineering and Cybernetics
GPA 4.8/5
Department of Radio Engineering and Cybernetics
GPA 4.8/5
Professional skills
Hard Skills
Statistics, Math, Machine Learning, Data Visualization, System Design
ML Skills
PyTorch, Tensorflow, XGBoost, LightGBM, CatBoost, PySpark, implicit, haystack, hugging face
Tool Skills
Python, FastAPI, Django, MongoDB, BigQuery, Airflow, PostgreSQL, git, docker, aws, ngnix, grafana, ci/cd,
Publications
The article provides an overview of recommender systems, focusing on factorization methods like Singular Value Decomposition (SVD) and its variants. It explores the evolution and types of recommender systems, emphasizing their mathematical aspects. The content is suited for readers with a background in machine learning, offering insights into the algorithms and concepts that drive these systems.
HABR
HABR
The article introduces GeoMLE, a new method for estimating the intrinsic dimension of data, particularly useful for non-linearly embedded, high-dimensional datasets. GeoMLE is efficient and accurate, utilizing geometric properties of data manifolds. It outperforms existing methods, especially in cases of non-uniform sampling and noisy environments, as demonstrated through experiments on various datasets.
Asian Conference on Machine Learning
Asian Conference on Machine Learning
The article presents a method for classifying brain diseases using brain network data. It focuses on overcoming the challenges of high data dimensionality and small sample sizes in connectome classification. The proposed method involves diagonalizing adjacency matrices to identify stable eigenstructures, which are then used for disease classification, particularly effective in detecting Alzheimer's disease. The approach outperforms basic methods and is comparable to advanced techniques.
Complex Networks & Their Applications VI
Complex Networks & Their Applications VI
