Education

Higher School of Economics, M.Sc. Data Science

Faculty of Computer Science, Moscow.
September 2022 – June 2024.
GPA: 4.80/5.00 or 3.80/4.00
Thesis: Tiny Language Models (Grade: 10/10)
- Trained 32 GPT-2-like models with varying hyperparameters and architectures.
- 3 models trained using Hugging Face.
- 29 models trained in pure PyTorch without using external libraries or pre-built models.
- Identified that training a single model per hyperparameter set is insufficient for meaningful comparison due to training randomness.
- Technologies: PyTorch, Hugging Face, Python, MLflow, LaTeX.
Taught in English.
Relevant courses:
- Data Science: Natural Language Processing, Introduction to Deep Learning, Research Seminar “Deep Learning”, Computer Vision, Machine Learning, Applied Machine Learning, Large Scale Machine Learning, Research Seminar “Data Scraping”, Algorithms and Data Structures, Python, SQL.
- Mathematics: Linear Algebra, Calculus, Probability Theory, Discrete Mathematics, Basic Statistics, Applied Statistics.

Higher School of Economics, B.Sc. Economics

Faculty of Economic Sciences, Moscow.
September 2014 – June 2018.
Minor: Data Science.
Taught in Russian and English.
Relevant courses:
- Data Science: Machine Learning, Modern Machine Learning Methods, Data Analysis, Applied Data Analysis Problems, Introductory Econometrics, Introduction to Programming.
- Mathematics: Linear Algebra, Calculus, Probability Theory and Statistics, Methods of Optimization.

Work Experience

MiraMedix

Natural Language Processing Engineer (Large Language Models).
January 2024 – June 2024.
Led a team of two natural language processing (NLP) engineers, managing task assignments and mentoring, while serving as the technical authority on all aspects of large language model (LLM) development.
Fine-tuned over 10 open-source LLMs for medical applications. Technologies: Hugging Face, PyTorch, Python.
Supervised data collection and scraping efforts from websites, books, and medical resources (e.g., International Classification of Diseases), providing technical direction and hands-on assistance to the team when needed. Technologies: Pandas, HTML, Beautiful Soup, Markdown, Requests, Apache Parquet, Hugging Face.
Developed guidelines for standardized data formatting and preprocessing after scraping.
Wrote all code for LLM fine-tuning and inference from scratch.
Led the setup and management of cloud infrastructure (Azure, AWS) for model fine-tuning and inference, initially handling most tasks and gradually delegating all operational responsibilities to subordinates while maintaining oversight.
Guided and supported team members with Linux, Git, SSH, Python, PyTorch, Hugging Face, and development environment setup and usage.
Organized Jira for project management and Confluence for documentation, writing detailed instructions and guidelines, and maintaining the team’s knowledge base.

Sber Automotive Technologies

Computer Vision Engineer (Deep Learning), Perception Team.
May 2022 – January 2024.
3D Object Detection: Increased mean Average Precision (mAP) by 9% through adopting a new model architecture, which also demonstrated superior performance in visual evaluations. Modified an existing ROS node (a component of self-driving car software) to integrate with the new model, enabling real-time object detection, debugging, and evaluation during vehicle tests in real driving conditions. Technologies: PyTorch, Python, Robot Operating System (ROS).
Lane Detection: Led the full development lifecycle of a lane detection component, from model selection and data preparation to training and ROS node integration into self-driving software. Code was merged into the main branch following approval by a cross-functional team of 8 engineers. Technologies: PyTorch, Python, ROS.
2D Object Detection: Trained a model from scratch on the internal dataset, improving both accuracy and processing speed. Technologies: PyTorch, Python.

Huawei Russian Research Institute

Computer Vision Researcher (Deep Learning), Intelligent Systems and Data Science Lab.
January 2022 – May 2022.
Contributed to the team that achieved the first place on the renowned nuScenes 3D object detection benchmark by experimenting with architectures, refactoring codebase, and developing custom support scripts.
Conducted weekly paper reviews and presentations, staying up to date with cutting-edge research; summarized and presented 1 research paper per week.
Technologies: PyTorch, Python.

Samsung Research Russia

Computer Vision Researcher (Deep Learning), Visual Algorithm Lab.
March 2020 – January 2022.
Extended Depth of Field Deblur project: Led the neural network development, designing and implementing all models independently, with a focus on GAN architectures. Technologies: TensorFlow, Convolutional Neural Networks (CNN), Generative Adversarial Networks (GAN).
Smart Autofocus project: Proposed the project, aimed at tracking objects and predicting their movements in three dimensions. Independently worked on prediction algorithms and collaborated with a team on depth estimation from phase autofocus data. Technologies: Kalman Filter, NumPy, PyTorch, Python, Long Short-Term Memory (LSTM).
Single-Shot Deblur project: Developed the codebase from scratch, trained the most successful model, wrote custom scripts for synthetic data generation, and participated in real dataset collection. Technologies: TensorFlow, Python, CNN, Bash scripts.

KPMG

Consultant, Financial Risk Management Department.
April 2019 – March 2020.
Developed and deployed models for corporate bankruptcy prediction for banks and insurance companies. Technologies: Python, Econometrics, statsmodels, scikit-learn, Pandas, NumPy, Matplotlib, Markov Chains.
Designed and developed NLP-based models from scratch to analyze online news and predict negative sentiment for specific companies. Technologies: Python, scikit-learn, Pandas, NumPy, Matplotlib.

Patents

Device and Method for Extended Depth of Field Imaging

2021.
International Publication Number: WO2023113193A1.
URL: https://patents.google.com/patent/WO2023113193A1/en.
Trained a neural network for extended depth of field imaging using the code I wrote from scratch.
Work completed during my time at Samsung Research Russia.

Skills

Deep Learning: PyTorch, Hugging Face, LLM, NLP, Computer Vision, TensorFlow, Keras, CNN, GAN.
Programming Languages: Python, C++, SQL.
Python Libraries: NumPy, Pandas, Matplotlib, SciPy, scikit-learn, statsmodels, NLTK, spaCy, Beautiful Soup, OpenCV, Requests, Seaborn, Pytest, PySpark, Pillow.
DevOps: Linux, Bash, Git, Docker, MLflow, Cloud Platforms (AWS, Azure), SSH, Vim, S3.
Robotics: ROS, Raspberry Pi, Nvidia Jetson Nano.
Management: Jira, Confluence.
Web: HTML, CSS, Jekyll, GitHub Pages.
Miscellaneous: LaTeX, Kalman Filter, Markdown, draw.io, LibreOffice, Microsoft Office.

Personal Projects

wiki.shlyapin.com

2024 – present.
Built a personal wiki to share my knowledge on various topics (Git, Docker, SSH, Markdown) with my subordinates at MiraMedix and the broader public. I plan to expand it with content on LLM, AI, Deep Learning, Data Science, Math, Computer Science, and more.
Technologies: Jekyll, GitHub Pages, Markdown.

Robot Car

2021.
Designed and built a robot car using Raspberry Pi, later upgraded to NVIDIA Jetson Nano.
Enabled remote control via a computer, with live video feed streamed directly to the controlling machine.

Additional Education

Coursera: Deep Learning Specialization (DeepLearning.AI), Mathematics for Data Science Specialization (HSE), Accelerated Computer Science Fundamentals Specialization (University of Illinois at Urbana-Champaign), plus 4 additional courses on Data Science and Computer Science.
DataCamp: 22 courses on Python, Deep Learning, TensorFlow, NLP, R, SQL, etc.
Other: Sololearn (C++), Codecademy (C++, Java).

Achievements and Awards

Samsung “Above and Beyond” Award “for practical and professional contributions to project outcomes”, 2020.
Second place in the United Traders trading competition among several hundred participants, 2015.
Prize winner at regional, municipal, and school stages of the All-Russian Economics Olympiad for schoolchildren, 2014.
Prize winner of the “High Proficiency” Economics Olympiad by the Higher School of Economics, 2014.
Prize winner of Saint Petersburg State University’s Economics Olympiad, 2013.

Additional Activities

Judge, Formula Student Self-Driving Cars Competition (Perception)

2022, 2023.
Served as a judge on the country level (Russia) for perception systems in the self-driving car category of Formula Student, an international competition with over 100 university teams from across the globe, evaluating object detection, sensor integration, and overall system performance.