Artem Vorozhtsov

Highly experienced in SE, ML, algorithms, and system design (14 years). Has profound ML & math background. I am hands-on, the one who "gets the job done" and launches impactful things. At Yandex: proposed and led deployment of first ML models for behavioural ad-targeting, trained and deployed ML models for ad CTR, conversion rates, models for behavioural targeting, shipped VCG auction. Proposed and led the transition to neural network-based feature extraction for ad machine learning models, developing user, ad, and placement embeddings using PyTorch. Integrated these embeddings into real-time CTR and CR prediction models, achieving substantial improvements in model accuracy and business profitability. Designed, implemented, and maintained a robust AB-testing system to validate model enhancements. At Yandex Market: proposed vision of using PyTorch models for demand prediction (dynamic pricing and replenishment), trained and deployed these models, that resulted in 10%+ in GMV. Led transition of ML pipelines from Catboost on CPU to PyTorch on GPU, made it more reliable. Reduced variance in the target metrics. Skills: ML: Classic ML (Catboost, XGBoost), Deep Learning (PyTorch, transformers, LLM, embeddings, text analysis) Python & Tools: PyTorch, scikit-learn, skopt, numpy, pandas, etc. Wolfram Mathematica: Expert. Data analysis, functional analysis, dynamic systems modelling, differential equations, regression, and classification ML tools, NetTrain. C++: Proficient in writing high-performance data analysis tools. Databases & Pub/Sub and Messaging Systems: Clickhouse, Hive, Kafka, industrial MapReduce systems & ML pipelines, Presto Programming Languages: Python, C++, Wolfram Mathematica, many dialects of scripting and SQLs languages. Analytics: Experience in designing A/B-test metrics and testing methodology for ad campaigns. PhD at MIPT, about combining heuristic a