• The process of processing Raman spectra (RS) was not automated, was performed in several programs and required a week of work from the researcher to obtain results.• The goal was to reduce the processing time of Raman spectra, automate the process, standardize for reproducibility, and combine all steps in one program.For this:• Developed a Python application for the full cycle of Raman spectra processing, which included automated data preprocessing, spectral decomposition, generation of new features and creation of machine learning models.• Implemented original algorithms for baseline correction and Raman spectra decomposition.• As a result, the processing process has become automated and standardized, increasing reproducibility and reducing time from a week to a few minutes.

• In a project to assess the probability of default on a consumer loan, I developed a full-fledgedservice based on data from a competition on kaggle (1.5 million entries).• At the exploratory analysis stage, it was possible to reduce the number of features from 488 to 190 without lossquality. For the models: CatBoost, XGBoost, LightGBM, hyperparameters were selected and ensemble models were created: Voting, Blending, Stacking, which increased the quality of the model to ROC AUC=0.8591, compared to Baseline (0.8531). The model was interpreted using SHAP.• The service is implemented using Fast API and Streamlit and deployed on the server using Docker.

• Automated business processes and optimized existing ones with an annual economic benefit of 6 million rubles.• Wrote/optimized more than 200 complex SQL queries ranging in size from 100 to 2000 lines.• Implemented changes to more than 100 technical specifications in non-standard 1C configurations.

He took part in the UMNIK Healthnet-NTI competition and won a personal grant to develop software for identifying and certifying the components of cell-tissue implants using the optical method during their production.The goal of the project is to create software for processing Raman spectra to solve problems in biomedicine.Developed software using Wolfram Mathematica and CUDA to speed up calculations. Using the software, results were obtained that are described in several scientific publications. The results of the work were presented at 11 international conferences, including the European Conference on Biomedical Optics 2019 in Munich.The results of the work have been published in several scientific publications, including articles in SPIE and Diagnostics.