Description

I help companies make their ML models faster, smaller, and ready for real-world deployment — on servers, mobile, or edge hardware.

At Huawei I compressed a computer vision model from 60MB to 1MB with 25x faster inference using quantization, pruning, knowledge distillation, and architecture modifications. At AVA I achieved 10x size reduction and 8x latency improvement on production NLP models.

I cover the full pipeline — profiling bottlenecks, applying compression techniques, validating on target hardware, delivering production-ready solutions.

Typical projects: model compression, computer vision optimization, inference speedup, PyTorch and TensorFlow development

Industry field of expertise

Languages

English
Fluent
Russian
Native or bilingual
Ukrainian
Native or bilingual
French
Basic

Workplace preferences

Can work on-site

Paris (up to 50km)

Huawei Technologies France
Machine Learning Engineer
TECH
July 2024 - January 2026 (1 year and 6 months)
Paris, France
Developed ML solutions across three projects focused on optimization and computer vision:

Compressed a computer vision model from 60MB to 1MB with 25x faster inference using quantization, pruning, knowledge distillation and architecture modifications — enabling deployment on resource-constrained hardware
Extended an optimization solver from convex to non-convex functions via custom transformations, achieving 3x faster convergence
Built probabilistic sequence forecasting models for app caching prediction using PyTorch
Pytorch Python Computer Vision model-optimization Deep Learning
AVA France
ML Engineer / R&D Intern
TECH
May 2023 - October 2023 (5 months)
Paris, France
Optimized a production speaker diarization model for edge deployment as part of master's thesis:

Reduced model size by up to 10x and latency by 8x using knowledge distillation, quantization, and pruning with minimal accuracy loss
Produced a full set of models covering different accuracy/latency trade-offs for flexible deployment across server and edge environments
Enabled potential deployment on mobile devices previously requiring server-side compute
Quantization Pytorch NLP TensorFlow Deep Learning
Fractal.AI
Data Scientist
TECH
December 2021 - October 2022 (10 months)
Kiev, Ukraine
Built an active learning system for time series outlier detection:

Designed a human-in-the-loop pipeline collecting real-time business user feedback on model predictions
Built Random Forest classifiers per time series enriched with sliding window and temporal features, retrained weekly on accumulated feedback
Enabled continuous model improvement adapting to domain knowledge that pure mathematical detection couldn't capture
Machine learning Deep Learning Python Time Series Anomaly Detection