Русский
Denis Drobyshev
Data science and machine learning intern
About
I work on machine learning and data analysis in Python. In 2026 I graduated from MIIGAiK with a bachelor's
degree in applied informatics, and during my studies I completed a retraining programme as a big data
specialist. My projects are on GitHub, and four of my libraries are published on PyPI. I am looking for a
data science internship.
Skills
- Python and SQL
- pandas, NumPy, matplotlib, SQL
- Machine learning
- scikit-learn, gradient boosting, PyTorch
- Statistics
- hypothesis testing, bootstrap, cross-validation
- NLP and CV
- embeddings, RAG, transformers, U-Net
- Tools
- Git, Docker, PostgreSQL, FastAPI, Linux
- Languages
- Russian (native), English A2–B1 (I read documentation and papers)
Projects
Choosing a business location from OpenStreetMap data
- Trained a gradient boosting model with a Poisson loss on 4,784 hexagons of Kaluga and 10 OpenStreetMap data layers.
- Evaluated it with spatial cross-validation: the 10% of cells with the highest prediction contained 98% of the city's pharmacies. Found and removed target leakage.
Reinforcement learning library, published on
PyPI
- Implemented 32 algorithms and environments for pricing, inventory, queueing and supply-chain problems.
- On 5 of 7 tasks the trained policy beat the classical operations-research method, and on the rest it performed on par (interquartile mean, 95% bootstrap interval, 5 runs).
Legal assistant for Russian law with citation checking
- Built hybrid search over six codes of law (BM25 and dense search with BGE-M3), with reranking and an NLI model that checks every citation.
- On the Civil Code (4,717 provisions) it reaches recall@5 = 0.92 and MRR = 0.94 on a test set of 18 questions. The service runs on FastAPI.
Semantic segmentation of aerial and satellite imagery
- Trained a U-Net without pretrained weights on the LoveDA dataset: mIoU = 0.438 on 1,669 validation images (urban 0.471, rural 0.379).
- Built a FastAPI web service that returns a georeferenced segmentation mask (GeoTIFF, GeoJSON).
Professional retraining
Development of geoinformation solutions using artificial intelligence methods
Dec 2024 — Dec 2025
MIIGAiK Digital Department, qualification "Big Data Specialist"
- Collected and prepared datasets, engineered features, tuned hyperparameters and evaluated models with cross-validation.
- Used linear models, decision trees, SVM, kNN, clustering, dimensionality reduction, ensembles and boosting.
- Trained neural networks in Keras: convolutional networks, transfer learning, autoencoders, GANs, YOLO object detection.
- Worked with spatial data (GeoPandas, Rasterio, OSMnx, Google Earth Engine) and integrated models into QGIS and Streamlit.
- Final project (Geomarketing_analysis): choosing a location for a fitness centre from OpenStreetMap data, 33 features on an H3 grid, RandomForest and GradientBoosting models, an interactive map.
Libraries on PyPI
- mlango: a framework for ML projects. Compares model versions and tests whether the difference is significant with an exact McNemar test.
- glia: a library for building LLM agents. 167 tests, code coverage of at least 90%.
- stadion: a benchmark for agents on operational problems. Compares an agent with the classical method and the exact optimum; the gap between those two ranges from 0.4% to 26.6% across six tasks.
Experience
Developer / DevOps
2026
MIIGAiK IT department (thesis project)
- Designed the database and backend of an academic journal portal and put it into production (Django, PostgreSQL, Docker).
- Set up sign-in through the corporate SSO and XML metadata export for bibliographic databases.
- Worked in the development team, agreeing technical decisions with colleagues and IT department staff.
IT support technician
Dec 2024 — Mar 2026
MIIGAiK IT department
- Administered computers in an Active Directory domain with SCCM and wrote automation scripts in PowerShell and Bash.
- Repaired and configured computer equipment.
- Helped users and, together with the support team, provided technical support for university events.
Production internship
2022 — 2024
MIIGAiK internal information systems
- Developed the interface of the student personal account and the core business logic of the university's internal systems.
- Wrote an SQL query that assembles the timetable from several related tables, for the university's Go services.
Education
Master's programme, applied informatics
since 2026
MIIGAiK, programme "End-to-end digital technologies in real estate management"
Bachelor's degree, applied informatics (09.03.03)
2022 — 2026
MIIGAiK, Department of Geoinformatics and Information Security. GPA 4.5 out of 5. Thesis: an academic journal portal (Django, PostgreSQL).
Achievements
- Author of lemma, a free 27-module course on ML, DL and RL.
- Fixes accepted into Gymnasium and Shimmy (Farama Foundation).
Other
- I enjoy cycling trips and fishing, and I read books.
← home