vdonin.com
RU

ML Engineer / Research Engineer · Moscow

Aleksei Vdonin

I build and ship NLP models: embeddings, rerankers, text classification and LLMs. I work across the full ML development cycle, from data preparation and training to evaluation and integration into production services. I also teach deep learning at the HSE Faculty of Computer Science.

Message on Telegram Download CV
Current role
Research Engineer, Machine Learning & Anti-Spam team
Teaching
Deep learning seminars, HSE Faculty of Computer Science
Education
HSE University, FCS — BSc in Applied Mathematics and Computer Science
MSc in Data Science, 2026⁠–⁠2028
Aleksei Vdonin

Skills

Languages
  • Python
  • C++
  • SQL
  • YQL
ML / DL
  • ML / DL (NLP, CV)
  • PyTorch
  • vLLM
  • Multi-GPU workflow
Data
  • NumPy
  • scikit-learn
  • Pandas
  • Polars
  • YTsaurus
  • PostgreSQL
Tools
  • Git
  • Linux
  • Jupyter
  • Claude Code

Experience

  1. March 2026 — present Current position

    Research Engineer

    Machine Learning & Anti-Spam team

    • Shipped embedding-size optimizations to the search service
    • Built the MVP of a new product
    • Trained a range of modified reranker models
  2. September 2026 — present Current position

    Deep Learning Instructor

    HSE University, Faculty of Computer Science

    • Teaching deep learning seminars to third-year students of the Applied Mathematics and Computer Science program and the Economics and Data Analysis program
    • Covering both the fundamentals of deep learning and an in-depth treatment of attention mechanisms and transformer architectures
  3. February 2025 — March 2026

    Analyst-Developer

    Yandex Alice ethics team

    • Developed and deployed ethics classifiers that raised the baseline quality of Alice's responses
    • Ran acceptance testing of new foundation models, evaluating the ethical quality of their responses with state-of-the-art methods
    • Introduced LLMs into existing processes, substantially reducing the person-hours required
    • Built numerous reusable tools and materials

    All tasks are described in deliberately abstract terms to comply strictly with the NDA: nothing that could even potentially fall under it is disclosed.

  4. September 2024 — February 2025

    Analyst Intern

    Emerging Information Security Methods Research group, Yandex Taxi

    • Developed and deployed an LLM-based classification algorithm that reduced manual workload by 15%
    • Applied state-of-the-art BERT- and graph-based methods to deduplication tasks
    • Supported and optimized existing business processes

Education

  1. 2026 — 2028

    Master's degree — Data Science

    HSE University, Faculty of Computer Science, Moscow

  2. 2022 — 2026

    Bachelor's degree — Applied Mathematics and Computer Science

    HSE University, Faculty of Computer Science, Moscow

Achievements

  • Finalist of the All-Russian Olympiad of School Students in Physics
  • Prize-winner of the HSE University student olympiad "Higher League" in Applied Mathematics and Computer Science (fourth year of the Bachelor's program)
  • Gold medalist of the International Experimental Physics Olympiad (IEPhO)

Projects

Bachelor's thesis

Unifying Tokenization: Finding the Optimal Strategy for Time Series

  • Built a PyTorch framework for comparing five time-series tokenization strategies on a shared Transformer backbone
  • Ran 90 experiments totaling 230 GPU-hours: three datasets, two forecasting horizons, three random seeds
  • Compared MAE / RMSE, training and inference time, GPU memory consumption and result stability
  • Automated the experiments, metric collection and analysis of accuracy-versus-compute trade-offs

GitHub Thesis (PDF, in Russian)

One of the best student projects of 2025 · HSE FCS

Document Recognition and Summarization in the HSE App X Mobile Application

  • Designed and trained summarizers based on custom architectures (including RNN + Attention) for integration into HSE University's official app, HSE App X, and the university's electronic document management system
  • Architectures were built for severely constrained environments: CPU inference within acceptable time (several architectures for different SLAs)
  • Selected as one of the best projects among Applied Mathematics and Computer Science students at HSE FCS in 2025

GitHub HSE website Report

NLP · NER

Criminal Case Research with Named Entity Recognition (NER)

  • Developed and trained a BERT-based model to extract key terms from court decisions issued under a specific article of the criminal code
  • Deployed an interactive web application for convenient use

GitHub Web app

DL · Transformer

DE → EN Translator on a Transformer Architecture, Written from Scratch

  • Implemented a Transformer model using basic PyTorch primitives
  • Trained on a single A100 with Weights & Biases logging and checkpointing
  • Ran dozens of experiments

W&B Report

Side project · linguistics

"Stress Dictionary" Browser Extension

  • A Chrome extension that automatically highlights Russian words on web pages and shows their standard stress in a tooltip
  • Fully open source; installs and runs natively in Google Chrome

GitHub

Interests

Padel, IPSC practical shooting, track driving (MRW), deep learning pet projects.

Contact