Artyom Kosarev
Data Scientist
11+ years exp
11+ years ML exp
C#
JavaScript
Python
SQL
Data Wrangling & Cleaning: 4/10
Predictive Modeling: 4/10
Reproducibility & Notebook Hygiene: 4/10
Active 2 days ago
Invite to interview
Message
Download CVCV
Overview
Technical skills
Timeline
Roles
Overview
A senior-level ML practitioner focused on feature-space optimization and custom GAN training with practical model-building strengths. The strongest proven skill is building a feature-set optimization engine as implemented in the BestFeatureGrid class which coordinates cross-validation, probability updates and multiprocessing. There is limited evidence of production hardening such as environment pinning, test coverage, CI, or deployment and monitoring pipelines.
Technical skills
C#
JavaScript
Python• Senior • 11y+
SQL• Senior • 11y+
Python
Django
Litestar
FastAPI• 5y+
Flask• 5y+
pySpark• 5y+
Databases
ClickHouse• 10y+
PostgreSQL• 7y+
Chroma• 4y+
Apache Kafka
Google BigQuery
AI/ML
Keras
Multimodal AI
OpenCV
RAG
TensorFlow
Time Series Forecasting
Spark
Deep Learning
Computer Vision
NumPy• 11y+
Pandas• 11y+
Scikit-learn• 11y+
SciPy• 11y+
PyTorch• 10y+
Statsmodels• 7y+
XGBoost• 7y+
LightGBM• 5y+
MLFlow• 5y+
Sentence-Transformers• 5y+
Transformers• 5y+
LangChain• 4y+
BERT
CatBoost
LLM
Vertex AI
NLP
Analytics
Seaborn
Matplotlib• 11y+
Plotly• 11y+
DevOps
Rest API
CI/CD• 11y+
Docker• 11y+
Git• 11y+
Grafana• 11y+
Nginx• 11y+
Prometheus• 7y+
GCP
Frontend
React.js
Redux
Timeline
Senior Data Scientist
•
Senior
East Mining Company
•
Full-Time
Developed and deployed a production coal price forecasting system for the Asian market, including 14-day-ahead predictions and improved model accuracy versus baseline. Built an automated data warehouse ingesting multiple external APIs with historical backfill and ongoing quality checks. Engineered an MLOps setup on GCP with BigQuery and Vertex AI to ensure reproducible experiments, and added a BERT-based NLP pipeline for weekly analyst report parsing to support forecasting features.
Python
SQL
PyTorch
Pandas
NumPy
SciPy
Scikit-learn
XGBoost
LightGBM
Google BigQuery
Vertex AI
GCP
Docker
Git
BERT
LLM
Senior Data Scientist
•
Senior
EuroChem Group AG
•
Full-Time
Built machine learning models to optimize equipment setpoints across a multi-stage chemical production line, improving yield without degrading quality. Designed and analyzed switchback A/B tests with confidence interval estimation and used unsupervised clustering to segment operating regimes. Integrated real-time Siemens PCS 7 SCADA data and performed regression analysis to relate sensor features to process yield KPIs.
Python
SQL
Scikit-learn
CatBoost
LightGBM
Statsmodels
SciPy
pySpark
Apache Kafka
PostgreSQL
FastAPI
Plotly
Docker
Git
Senior Data Scientist | Data Engineer
•
Senior
Roxleg
•
Full-Time
Served as tech lead for an ML-based algorithmic trading platform, including microservice API design and an automated experimentation workflow to speed up hypothesis testing. Worked on automated stock portfolio management for market brokers as a founding project. During company pivot and consulting work, led data pipeline and production deployment efforts such as migrating DWH to ClickHouse, building RAG-based chat automation, and automating claims and robotics CI/CD.
Python
SQL
Pandas
NumPy
SciPy
Scikit-learn
LightGBMsince 2021
Statsmodels
PyTorch
Transformers
LangChain
Sentence-Transformers
Chroma
FastAPIsince 2021
Flask
pySparksince 2021
MLFlow
ClickHouse
PostgreSQL
Docker
CI/CD
Git
Matplotlib
Plotly
Grafana
Prometheus
Senior Data Scientist
•
Senior
IBM East Europe / Asia
•
Full-Time
Developed models to predict lithological transitions ahead of drilling using sensor telemetry and mathematically engineered features from raw data. Implemented heavy ETL pipelines and enabled real-time inference during drilling operations. Built additional regression models for well productivity and optimized feature selection and custom metric computation to support timely evaluation and client stakeholder reviews.
Python
Scikit-learn
XGBoostsince 2019
NumPy
Pandas
SciPy
Statsmodelssince 2019
PostgreSQLsince 2019
Matplotlib
Git
Docker
Data Scientist | Teamlead
•
Lead
JetStyle
•
Full-Time
Created a deep-learning anomaly detection approach for industrial equipment maintenance using a seq2seq autoencoder combined with FFT-based spectral vibration analysis for earlier malfunction detection. Built a high-load data engineering pipeline and deep-learning ETL workflow for sensor streams, including sliding-window preparation and feature engineering. Drove delivery via industry presentations that led to pilot deployments in multiple industrial facilities.
Pythonsince 2015
SQLsince 2015
ClickHousesince 2015
Pandassince 2015
NumPysince 2015
SciPysince 2015
Scikit-learnsince 2015
PyTorchsince 2015
Dockersince 2015
Nginx
Grafanasince 2015
CI/CDsince 2015
Gitsince 2015
Matplotlibsince 2015
Plotlysince 2015
Ural Federal University
Bachelor's Degree •
Information Technologies
Full Stack Engineer
•
Middle
JetStyle
•
Full-Time
Built backend components for an indoor navigation system based on Wi-Fi data. Implemented backend services for geophysical telemetry visualization and developed a mobile interface for B2B billing personal accounts. Contributed to end-to-end application infrastructure with containerized workflows and CI/CD automation.
FastAPI
Flask
React.js
Redux
Git
Docker
CI/CD
Middle Data Scientist
Confidence: Medium ML Practitioner
A senior-level ML practitioner focused on feature-space optimization and custom GAN training with practical model-building strengths. The strongest proven skill is building a feature-set optimization engine as implemented in the BestFeatureGrid class which coordinates cross-validation, probability updates and multiprocessing. There is limited evidence of production hardening such as environment pinning, test coverage, CI, or deployment and monitoring pipelines.
Statistical Rigor
3/10
Correct use of statistics
Some sound evaluation practices (cross-validation and a scorer abstraction) are present but there are no explicit assumption checks, uncertainty quantification, multiple-comparison controls or causal analysis; statistical rigor is limited to cross-val mean scoring.
Data Wrangling & Cleaning
4/10
Preparing and cleaning data
Practical dataset loading and preprocessing for images is implemented with a preload option and on-the-fly resizing, but data-cleaning and robustness (missing file handling, modern image IO, input validation) are thin and use deprecated functions.
Exploratory Analysis & Visualization
2/10
Exploring and visualizing data
Very light visualization and exploration: training loop prints shapes and saves generated images but there is no structured EDA, written interpretation, or dashboarding artifacts.
Predictive Modeling
4/10
Building models that predict
Modeling work includes a custom GAN architecture and a nontrivial feature-space optimizer that deep-copies estimators, uses cross-validation, refits models for importances and orchestrates multiprocessing; error analysis, calibration and richer model diagnostic tooling are limited.
Business Insight & Impact
1/10
Turning analysis into business value
Little to no demonstration of connecting analysis to business metrics, cost-of-error reasoning or actionable recommendations; problem framing is technical rather than business-driven.
Reproducibility & Notebook Hygiene
4/10
Clean, repeatable analysis
Reproducibility practices exist at a code level: seeded random states per process, result persistence via shelve and deterministic random_state construction, but environment pinning, tests, CI, data versioning and packaging are absent.
Verified artifacts
Expertise
Analytics• Middle
Technologies
Deep Learning
Seaborn
Matplotlib• 11y+
Computer Vision
Plotly• 11y+
Litestar
Models• mentioned only
Recommendations
- Develop feature-selection and model-evaluation components, expanding BestFeatureGrid into a reproducible library with unit tests and CI.
- Prototype and iterate on custom GAN research experiments and training pipelines, adding training telemetry and systematic error analysis.
- Harden data-loading and preprocessing: replace deprecated image IO, add input validation, missing-data handling and unit tests for dataset code.
- Package and operationalize models and optimizer workflows with environment pinning, reproducible run scripts and basic deployment/monitoring hooks.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories: