Data Scientist

Product: NLP-based customer feedback analyzer. Worked on NLP tasks for analyzing customer reviews (topic and sentiment classification). Researched SOTA approaches and deployed ML models to production. Improved classification accuracy by fine-tuning models on real-world production errors. Automated data collection and preprocessing pipelines; set up model inference and monitoring. Tools: Python, PyTorch, nltk, Hugging Face Transformers, scikit-learn, FastAPI, Docker, MLflow, spaCy, pandas, Git, Peft, Trl, Unsloth, Deepspeed, Pymorphy.

Product: Developed and deployed ML solutions for external clients. Delivered two production-grade projects in the areas of Computer Vision and Time Series Forecasting. Built segmentation and classification models to analyze customer review photos (identifying and recognizing product categories). Introduced a multimodal approach by combining image and text analysis to improve classification performance. Trained and fine-tuned models such as RegNet, DenseNet, ViT, and ConvNeXt on noisy real-world data. Improved model performance significantly (Segmentation DSC: 0.70 → 0.87, Classification ROC AUC: 0.85 → 0.94) through retraining on production failure cases. Handled end-to-end pipeline: data sourcing, annotation setup, model training, and deployment. Analyzed 3–5 years of historical time series data (e.g., stock prices, financial indicators) to identify trends and build predictive models. Conducted research and adapted forecasting solutions to meet business goals. Deployed models into production with monitoring and retraining mechanisms. Tools: PyTorch, OpenCV, timm, PIL, Hugging Face Transformers, PEFT, TRL, Unsloth, DeepSpeed, NLTK, pymorphy2, NumPy, scikit-learn, FastAPI, MLflow, Docker, Prophet, statsmodels, Plotly.