Web-site for playing hard intellectual games in Russian.Using internet is allowed when solving tasks.~50 users play regularly (every few weeks) with a cost of ~$10 for a team of ~3-5 people.I develop website and create most of the games.Web-site can check the answers and show the current results.Technologies: Python, Django, AWS

Developing ELT procesess for Analytics Datastore with TBs of data, optimising materialised views.Deploying LLM models on Snowpark Container Services (link to my article in Snowflake Medium: https://medium.com/snowflake/using-snowflake-for-ai-completion-in-manychat-37bfaa914f8c)Technologies: Python, Snowflake (including Snowpark and Cortex), SQL, Docker, huggingface, pytorch.

Researched experiment trackers (neptune.ai/WandB/ClearML/MLFlow), deployed neptune.ai and developed Python wrapper for researchers to track pipeline runs.Developed several features of a big data Prefect pipeline: cache storing, tests, new stage proposed by a researcher.Technologies: Python, neptune.ai, Prefect, Gitlab CI/CD, Kubernetes, Docker.

Main project: Shortly after joining the team was promoted to lead. I was leading the team of 5 people developing model lifecycle management system for a client company (top-3 telecom in Russia).Work with a client included audit, system design, monthly developing a list of tasks.The main goal of the system is to track the stages of a model and decrease Time-to-Market.System integrated with Jira, Confluence, Gitlab, MLFlow, Airflow, Oracle, Hive.Technologies: Python, PostgreSQL, Docker, Fast API, Camunda, Redis, Kafka, Node.js, ReactSide project: Developed a prototype of A/B-platform for the GlowByte platform.Technologies: Python, PostgreSQL, FastAPI

Developed a data process advertising mobile app to users with similar vectors to users who got deep conversion event in this app.Developed a data process advertising mobile apps to users visiting urls connected to app.Result: ~5% increase of clicks in a mobile app segment.Developed new features for a e-comm ads model.Result: ~5% increase of clicks in a e-com segment.Technologies: Python, C++, yt/hadoop, MapReduce, Graphana, ClickHouse, Spark (custom analogue), gRPC,MatrixNet/CatBoost, Word2Vec