SUMMARY

Senior machine learning engineer with 10 years of experience in computer vision, scalable data pipelines, and deploying production-grade deep learning models. Expert in Python, proficient in C++, with strong foundations in PyTorch, ONNX, multi-processing, and asynchronous programming. Focused on research-driven model evaluation, custom metric development, and building tools to improve model robustness and quality. Experienced in hybrid and remote teams, with open-source and competitive machine learning contributions.

TECHNOLOGIES

LanguagesPython (expert), C++ (proficient, CMake, Conan), Rust (working knowledge)
Machine LearningPyTorch, TensorFlow, NumPy, Pandas, Dask, Jupyter, Streamlit, altair, scikit-learn
Data EngineeringAzure DevOps, GitHub Actions, SQL, Weights & Biases, AWS S3
Othermulti-GPU training, model quantization (fp16, int8), software design patterns, data structures, algorithms

WORK EXPERIENCE

Senior Data Engineer

CARIAD SE (Hamburg, Germany)

2024—present

  • Joined a team of senior embedded systems developers, contributing to the deployment of triggers and learning modern development practices, including C++20, CMake, Conan, and MicroPython.
  • Diagnosed and resolved issues with a Docker dev container on macOS, improving the robustness and reliability of the development environment.
  • Developed a utility that generates and sends protobuf messages to an AI image-triggering subsystem, enabling IPC communication with the tool typically receiving images from an embedded camera to simplify debugging.
  • Redesigned the object detection triggering model to support multiple loss heads for entropy-based triggering used in the data collection fleet.

Senior Computer Vision Engineer

CARIAD SE (Hamburg, Germany)

2023— 2024

  • Within 1.5 years, became the top contributor (by LOC) to PyTorch-based model training codebase in a team of 8 developers working on AI triggers for intelligent data collection. Built five new modules, including model evaluation, image triggering, visualization, logging, and CLI functionality.
  • Led the team in contributions to the test codebase, accounting for 25% of surviving LOC changes and achieving 95–98% coverage for newly added functionality. Introduced a rigorous TDD approach and decoupled data types and interfaces to streamline testing.
  • Co-authored a Bounding Boxes Localization Uncertainty Estimation algorithm that accurately identified poor-performing images to automatically include them into subsequent Active Learning iterations.
  • Designed and implemented a flexible model evaluation framework for segmentation and object detection models. Based on researched papers, introduced two new object detection metrics (variants of mAP) and refactored a legacy monolithic evaluation class into four modular components, enabling easy interactive use and significantly improving testability.
  • Refactored the model evaluation and reporting pipeline, enabling the team to run 1,000+ tracked, reproducible, and reliably measured experiments within a year, by providing essential tools and integrations.
  • Created five well-documented Jupyter notebooks with interactive reports to deliver prototyping and analytics for a team of 8 senior ML engineers and the team lead. These reports facilitated critical decisions on evaluation strategies and metrics.
  • Developed three flexible data-splitting strategies for computer vision datasets, widely adopted by team members for various projects.
  • Introduced Altair and Vega-Lite for data visualization. Developed a new violin plot definition for the team's W&B dashboard, now used as a standard plot for training experiment visualizations.
  • Diagnosed and fixed four critical issues in a third-party evaluation framework (TensorFlow-based), including incorrect class mapping and a subtle file-ordering error. These fixes ensured reliable evaluation results for object detection models.
  • Developed a versatile set of tensor manipulation utilities, reused in six different modules across the codebase, streamlining operations and reducing code duplication.

Staff Data Scientist

Blue Yonder Group, Inc. (Hamburg, Germany)

2020—2023

  • Collaborated in a team of five core contributors on a large-scale project for a major US grocery chain (2000+ stores). Solely developed three critical machine learning pipeline components–data preprocessing, model training, and forecasting–which contributed to achieving a 100% daily data delivery success rate during the first year of service.
  • Developed a library for distributed categorical variable encoding, capable of processing up to two billion records per run. Built with the Dask framework, it ensured reliable weekly execution for three large-scale customers, achieving a flawless operational record.
  • Co-developed a forecasting library with a unified API, enabling to run models developed both in Python and Java, in interactive and batch modes. It became a core component of the new forecasting platform.
  • Extended an in-house developed Generalized Additive Modelling (GAM) library by adding price elasticity modelling functionality, used to optimize pricing for one of the largest and longest-standing customers.
  • Migrated core components of the model training pipeline from a custom in-house solution to the PyTorch Lightning training loop. Refactored existing functionality to leverage PyTorch standard interfaces, including callbacks, datasets, and data loaders.

Machine Learning Engineer

Smaato, Inc. (Hamburg, Germany)

2019—2020

  • Played a key role in a team of six engineers, co-designing and building a Python framework unifying data management and model training processes, serving as the core machine learning solution for a company of 200+ employees.
  • Built a scalable machine learning pipeline capable of training deep learning models on up to 300 GBs of ad bidding requests data, utilizing PySpark, Databricks, and AWS SageMaker.
  • Wrote comprehensive documentation for the developed package, covering the API, design choices, capabilities, and limitations. This documentation became a core reference for a team of 10 engineers and data scientists.
  • Facilitated the transition from TensorFlow 1.x to TensorFlow 2.x with eager execution and the Keras API, improving code readability and maintainability. The migration addressed performance issues caused by incorrect handling of training flags during inference, resulting in improved model performance.

Software Engineer

Self-Employed Contractor (Remote work)

2014—2019

  • Collaborated with over 20 clients on diverse projects, including building machine learning models, automation scripts, data processing pipelines, APIs, and desktop applications.
  • Developed a configurable machine learning engine with analytics and reporting subsystems, processing millions of e-commerce transactions per client for B2B use cases.
  • Designed and built a backtesting framework for validating commodity trading strategies.
  • Created API endpoints and ORM models for a predictive food consumption system integrated with a smartwatch.
  • Ported and enhanced an OpenVPN-based Windows VPN client for macOS, improving cross-platform functionality.

PROJECTS

Kaggle — Competitions Expert

2019—2024

EDUCATION AND CERTIFICATIONS

C++ Development Course, Udacity

2025 (ongoing)

Object Tracking Course, Udacity

2024 (ongoing)

AI and ML Development Courses, Udacity

2017

Software Engineering Diploma, Surgut State University

2009—2014

Equivalent to a Master's degree in many European countries

INTERESTS

Exploring new languages—both programming and natural. Currently, learning German and improving skills in C/C++. Enjoy building applications in Rust, developing simple tools in C, and experimenting with GenAI and web technologies. Occasionally participate in Kaggle competitions to explore new models and enhance data science and machine learning expertise.

Tip: press Ctrl+P or Cmd+P to display a print-ready page (works best in Chrome)