Senior AI Engineer with a track record of taking agentic and ML systems from prototype to production. Built the agent architecture, shared platform and evaluation harness that the Novartis finance data-science department builds its agents on, after delivering ML forecasts covering €20B+ in yearly costs across 30+ countries. Outside work, built and launched production multi-agent apps and open-source agent evaluation tooling.
Earlier: Transfer Pricing Consultant, PwC (2021) · Teaching Assistant, Statistics & Microeconomics, HSE (2019–22)
Also: PasteReady — agentic document assistant that reads email and documents and triggers workflows such as hotel and flight check-ins or ID renewals · Hoodwise (neighbourhood scoring) · ROIInvest (property analytics).
Agentic AI & LLMs: LangGraph, LangChain, OpenAI Agents SDK, MCP servers (FastMCP, OAuth 2.1), agent evaluation (LLM-as-a-judge), agent memory, RAG, AWS Bedrock, OpenAI / Anthropic / Gemini SDKs, LoRA fine-tuning, vLLM
ML: Time-series forecasting, gradient boosting (XGBoost, LightGBM, CatBoost), NLP, PyTorch, scikit-learn
Engineering & MLOps: Python, TypeScript, React, FastAPI, Docker, Jenkins, GitHub Actions, unit/integration/E2E testing, AWS (Lambda, ECR, S3, DynamoDB, SageMaker, Step Functions), GCP Cloud Run, Terraform, MLflow, Langfuse
Data: SQL, Snowflake, Databricks, PySpark, Polars
Languages: English (fluent), Russian (native), Spanish (intermediate)