LEAD AI/ML ENGINEER at DeepPavlov (2023-10 – 2026-02)
Led a team of 6 engineers to launch a new commercial division, bridging foundational AI research with the delivery of enterprise‑grade, user‑facing products.
- Standardized AI development practices via DevContainers and established GitLab CI/CD pipelines, automating Kubernetes deployments to transition ML research into production‑grade enterprise services.
- Designed a stateful, multi‑agent conversational system for automated test generation, leveraging SSE to enable human‑in‑the‑loop validation for complex UI/API scenarios.
- Architected an agentic legal assistant utilizing multi‑step reasoning chains and a RAG pipeline to automate document risk assessment for a Top‑3 Russian bank, reducing manual review latency by 70%.
- Optimized inference for Encoder‑Decoder Language Models by engineering a custom engine, achieving 10x latency reduction via torch.compile, int8 quantization, and custom KV‑caching.
- Directed the full model development lifecycle, leveraging scaling laws and proxy models for cost‑effective pre‑training (e.g., SAM) and building post‑training pipelines for SFT, PEFT (LoRA), and preference alignment (DPO, GRPO).
SENIOR AI ENGINEER at TIMG (2022-09 – 2023-10)
- Architected an AI‑driven search autocomplete engine to handle spelling errors and out‑of‑vocabulary queries. Engineered a Char‑CNN and LSTM pipeline, improving search relevance by 60% within strict sub‑100ms latency budgets.
- Orchestrated an event‑driven ticket routing and real‑time monitoring platform, fine‑tuning RoBERTa model for context‑aware segmentation, combined with CDC and RabbitMQ delayed messages for state tracking, reducing average response times by 40%.
MACHINE LEARNING ENGINEER at Sberbank CIB (2021-12 – 2022-08)
- Deployed an asynchronous NER routing system using Kafka, ensuring zero‑loss event processing during traffic spikes and reducing email routing time by >90%.
- Launched an automated forecasting chatbot, cutting analyst workload by 75%. Combined NLP intent parsing with FAISS vector search to retrieve datasets, triggering dynamic XGBoost model training to serve cached predictions and custom visualizations.
HEAD OF ANALYTICS & DATA SCIENCE at TIMG (2011-12 – 2015-04)
- Founded the analytics department, architecting ETL pipelines to a PostgreSQL Data Warehouse to centralize BI and protect OLTP databases from read contention.
- Mentored a data team to stabilize the supply chain, developing demand forecasting models that reduced lost‑to‑sales ratios by 30% and import costs by 15%.
