- Trained a generative model (llama-2, mistral, mixtral) for generating texts on user demand, applying SFT LLM methods (7b/13b) including PEFT techniques (p-tuning, LoRA). Implemented G-Eval metric which increased from 3.2 to 4.4, also increased quality on copywriter markup by 1.3 conditional units- Designed and implemented a ranking system (BERT) for LLMs using (p-tuning, LoRA), which raised the text generation quality score from roc-auc = 0.67 to roc-auc = 0.87- Designed and implemented AI chatbots using LLMs and OpenAI API, providing conversational interfaces for websites and PDF documents. Utilized LangChain library to create and optimize NLP pipelines, improving chatbot performance and user interaction- The obtained model consisted of several stages: the first stage - Faiss+Cosine Sim+RoBERT, the second stage - fine-tuned CatBoost, XGBoost, LightGBM- BERT model were used to extract embeddings from the text. Based on the results of experiments, I chose RoBERT, which provided an average increase in Precision@k by 0.15 compared to other models- For the classification task, I trained (XGBoost, CatBoost, LightGBM + Bagging) and reduced it to a matching task, increasing Precision and PR-AUC scores by about 0.12Stack: pytorch, transformers, BERT, CatBoost, LightGBM, PEFT, loralib, kserve, Docker, FastAPI, Faiss

- Implemented SEO and SEM strategies, achieving over 352% ROI through data-driven decisions- Improved conversion rates on multiple websites by up to 41% through user behavior analysis and dashboards- Built a sentiment analysis model using classic ML on Trustpilot reviews to predict customer satisfaction across platforms and detect insights- Utilized Google Analytics and Python for data analysis and automated reporting, enhancing performance tracking- Collaborated with marketing, product, and IT teams to integrate data insights into strategic decisions- Worked extensively with MySQL and PostgreSQL databases for data management and analysisStack: Python, Pandas, Google Analytics, Scikit-learn, MySQL, PostgreSQL, XGBoost, Logistic Regression,NLTK, Matplotlib, Seaborn