AI Research Engineer | ML Engineer
AI Research Engineer with ~4 years in ML/AI, focused on agentic systems that hold up in production.
At Raiffeisenbank, I architected a multi-agent AI platform from scratch: autonomous agents generated product hypotheses, and 67% of them were accepted by product teams for testing. Earlier, I built an A/B testing framework whose experiments delivered a $25k monthly revenue uplift. Today I work on both sides of the field. At AIRI, I develop agentic VLM-based RAG for document understanding, currently outperforming published SOTA baselines. At HSE, I mentor graduate students in Advanced NLP, from Transformer internals to multi-agent systems. Before AI, I built my career in aviation rising from Maintenance Engineer to Production Manager of 60+ engineers in safety-critical operations. That is where my production-first mindset comes from: an AI system is only as good as its behavior under real-world constraints. Technical Stack: • Agentic & LLM: Agentic Workflows, LangChain, LangGraph, MCP, RAG, vLLM, Transformers • ML & Evaluation: Python, PyTorch, scikit-learn, Ragas, Experimental Design, A/B Testing, Drift Monitoring • MLOps & Orchestration: Apache Airflow, Kubernetes, Docker, CI/CD, MLflow • Data & Databases: PySpark, SQL, Hadoop, Vector DBs (Milvus), MongoDB, AWS S3 • Backend & Tools: FastAPI, Litestar, Kafka, RabbitMQ, Redis, Bash, Linux, Git, Pytest Open to roles: AI Research Engineer • LLM Engineer • Agentic AI Engineer • Applied Research Scientist
AIRI
May 2026 — PresentDeveloping VLM-based RAG for financial-document QA; reports outperforming published baselines on T2-RAGBench, with a first-author paper in preparation.
• Designing an agentic VLM-based RAG system for text-and-table document QA over real-world financial reports, outperforming published SOTA baselines in numerical-reasoning accuracy (Number Match) on T2-RAGBench; first-author conference paper in preparation.
Higher School of Economics (HSE University)
Russian university.
Sep 2025 — PresentSupports graduate NLP teaching for courses with 300+ students and supervises hands-on LLM implementation and fine-tuning.
• Teaching assistant for graduate Basic & Advanced NLP courses (300+ students); designed curriculum and examined 30+ students on topics from Transformer internals to Multi-Agent Systems. • Supervised low-level LLM implementation - manual Qwen2.5 architecture (RoPE, RMSNorm, SwiGLU) for speculative decoding, plus PEFT/QLoRA fine-tuning and GPTQ quantization.
Raiffeisenbank
Banking institution.
Jun 2024 — May 2026Built a production multi-agent platform with 67% product-team acceptance of generated hypotheses, and delivered labeling, batch ML, and HR-policy RAG systems with measurable throughput and latency gains.
• Architected and delivered a corporate multi-agent AI platform where autonomous agents explore user data to generate product hypotheses, achieving a 67% acceptance rate from product teams for further testing. Equipped agents with an Agentic RAG tool (FastMCP) for banking-domain context and a Zero-Touch QA agent (Ragas) for automated evaluation, served by an internal vLLM stack (GLM for agent reasoning, Qwen-2.5-Coder for code generation) in isolated Micro Sandboxes.
Sportmaster
Retail company.
Jan 2023 — Jun 2023Built a production classification pipeline to identify warehouse safety risks.
• Built a production ML classification pipeline for automated warehouse safety-risk detection.
VseInstrumenti.ru
Online retailer of tools and home-improvement products.
Jun 2023 — Feb 2024Led data science for three products and ran 40+ A/B tests, contributing to a reported $25K monthly revenue increase.
• Led Data Science for three core products on a 12M+ MAU platform; ran 40+ online A/B experiments with advanced variance reduction (Stratification, Delta Method), directly driving a $25K monthly revenue increase.
Higher School of Economics (HSE University)
2023 — 2025Saint Petersburg State University of Civil Aviation
2016 — 2021