Data Science Specialist
Tools:- Python, Git, Spark, AWS, Azure, Gurobi, PytorchWorked in R&D group which was researching application of fine-tuning large language models for client specific tasks:o Trained NLP models on GPUs on cloud (GCP Vertex AI, AWS Sagemaker).Created digital marketing personalization engine for large retailer:o Customized first level ALS matrix factorization recommendation engine.o Recommendations for 10m+ customers on Azure cloud stack.o Implemented CI/CD using Azure DevOps pipelines, Databricks workflows and kedro library.o Optimized Spark queries that speed up calculations from several hours to several minutes.Developed rail freight delivery optimization model for a wagon park owner. Client has 60k wagon park and more than 150k deliveries every month. Task was to increase rail carriages utilization while minimizing cost: o Developed from scratch integer linear programming model using Gurobi package.o Achieved improvement in wagons park utilization from 87% to 94%.Created model for SKUs pricing elasticity estimation for a big retailer. Client has hundreds of stores and tens of thousands of different SKUs. Task was to achieve better pricing for non-KVI products:o Multiple time series forecasting of product store sales.o Did A/B tests with covariates adjustments.o Solution showed $10m+ improvement in sales if using optimal price policy.Applied classical Machine learning for different stages of oil refinery process. Worked with Fluid Catalyst Crackers and Hydro Catalyst Crackers:o Usually, you have 6-8 time series models for each of the factory final products.o On top of the product models, I’d created revenue optimizer.o Potential increment in revenues was up to $15m.
