Andrey Tyunyatkin

• Built and scaled the AI Product Discovery process: ML Growth Point board and templates, ML task grooming, ML task prioritization system, the process of working with the "bloating backlog", Gantt-based task decomposition. • Assembled and led a high-performing Data Science team, establishing ML Delivery best practices: research-repository, ClearML for experiments and production-models monitoring, Model Registry, materialized Feature Store (and OMD data catalog), devcontainers, code review and model review, checklist systems, model deploy with dynamic retraining and dataset DQ-checks, MapReduce for resource-intensive computing. • Led the development and deployment of pricing models, driving an additional $2M GMV without compromising seller satisfaction (measured via CSI & NPS). • Managed the development of Content Delivery models to the prod, achieving a 30% increase in CTR. • Implemented an AI-linter that improves the speed and quality of code reviews across multiple teams.

• Role of an "AI Evangelist" in the Compliance Department (over 600 employees): Developed AI solutions for anti-money laundering and internal investigations. • Conducted over 50 in-depth interviews and compiled a backlog of 100 ML growth points across 6 departments. • Led the development of a payment classifier, achieving a financial impact of up to $5 million annually (Precision and Recall above 0.9). • Managed the development of a data drift monitoring solution, resulting in the reduction of up to a $1 billion in suspicious turnover by promptly responding to data changes. • Implemented MLOps for 3 ML teams, including Feature Store, checklist systems, agile methodology, research workflows, AI utilization, clean code practices, model review and testing, and (mutual) employee training. • Created and managed an internal library with ML and autoML solutions for data analysts. • Established knowledge sharing practices: from knowledge bases to grooming sessions and hackathons. • Contributed to the development of anomaly monitoring in banking reports (time series analysis). • Optimized the architecture for ML model delivery, focusing on uptime and ML model observability.

I lead a project on self-driving (computer vision) - more than 20 models and drivers have been developed. For example: a detector of distracted driving, a pedestrian tracking system, a model for detecting road signs. I managed to increase the project budget by 2 times due to the stabilization of the development process and ensuring the promised results. Requirements: - Product management: I conduct customer development sessions, design CJM and UI/UX. - Project management: I write the annual project plan, set the regulatory requirements, manage the development team and conduct the brainstorms. - Tech Lead: I conduct pair programming sessions, choose the stack, solve ML tasks (computer vision, neural networks, ensemble machine learning methods), and test the system. Stack: PyTorch, sklearn, YOLO, ResNet, Faster STARTUP, Faster AVI, Celery, PyDoc, Gunicorn, Nginx, Docker, Ubuntu.

I led the end-to-end development process for more than 10 products. The result is an ecosystem of AI services. The project at different stages involved from 7 to 20 developers. Responsibilities: - Product management (from in-depth interviews to PRD/CJM and product vision formulation), - Project management (flexible methodologies and principles of project management in conditions of high uncertainty), - Tech leading. Stack: FastAPI, Flask, PyTorch, sklearn, SpaCy, Stanza, SBERT, KeyBERT, ruBERT-sentiment, DeepPavlov, CNN-LSTM, RMN, Facenet, Geogebra API, React.js, PostgreSQL, MongoDB, ClickHouse, Superset, Gunicorn, Nginx, Docker, Docker Compose, GitHub CI/CD, Ubuntu. Result: - The team implemented 6 USE simulator generators, two widgets for training students and a task generation system. More than 60,000 unique tasks have been generated. - Together with the team, I developed a system for predicting grades and competencies of students: f1/acc/prec/rec: 0.922 / 0.94 /0.94 /0.94 . Implemented Superset dashboards. - Supervised the creation of a content indexing system, including telegram chats (conversation topics, emotional coloring, harassment). I designed a unified interface and dashboard system. - Designed and supervised the development of a proctoring service (body position, face, eyes; analysis of background sounds) with the ability to monitor the exam in real time - I Mentored the team developing chatbots for monitoring the educational process - For testing and training models, I managed the development of a data synthesis system for the educational process (demonstrations, Syntetic Balancing) The developed products have passed alpha testing and are awaiting integration. I successfully mentored new employees: there are two cases of growth of my subordinates to the team leads of other teams.

Project Lead on the side of Rebels.ai in a project of creating a recommendation system for assigning training courses to sellers. Worked within the client's team under the mentorship of Rebels.ai CEO. Responsibilities: - Participation in workflow optimization - Conducting meetings - Creating and holding presentations (for other departments and stakeholders) - Data delivery pipelines - ML models: architecture, training, boosting - Analytics and metrics Delivered: - Data delivery, aggregation and analysis pipeline, as well as interactive product metrics dashboards (Clickhouse, PostgreSQL, Superset) - A system for assigning training courses to sales managers (AUC up to 0.9) inside a fully configurable pipeline (VowpalWabbit, CatBoost, sklearn, Watertape) - Easily-understandable product metrics (client's demands: centering, normalization and discreteness - like examination points)

As an ML engineer at Rebels.ai have been transfered to a project of developing an automated school test generation system - Natural language processing (NLTK, Stanza, FuzzyWuzzy) - Text lemmatization. Created a corpus (more than 1 GB) of Russian literature - Created several test generators - Used the text corpus to generate test tasks The result of work is in a private repository, NDA.

Research in areas of: discrete mathematics, scheduling theory, interpolation methods, the problem of minimizing the maximum lateness. Published 6 articles, 5 reports at conferences, 3 abstracts. Twice received the best report award. - Conducted several numerical experiments to confirm the effectiveness of the dual algorithm for solving the problem of minimizing the maximum delay. - Formalized the algorithm of the interpolation approach to solving the scheduling theory problems. - Derived two new polynomial cases of the problem of minimizing the maximum lateness and presented an algorithm of the interpolation approach based on these cases.

Automated and manual testing of Solidity smart contracts. Report preparation.

Managed the development team at a startup related to the analysis of the rental housing market. - Started working under supervision of an experienced mentor - Formed up a team of 8 developers - Set up a workflow by implementing Agile methodologies - Together with the team implemented a system for parsing and storing data on the rental housing market (Python: Flask, BeautifulSoup, RegExp, etc., Heroku, PostgreSQL). 12 parsers using proxy servers - Together with the team, I implemented a Django server for analyzing and visualizing the parsed data - Wrote scripts for analyzing social media posts

Department of Control Methods