Professional Summary
ML engineer with 5+ years shipping models to production and building LLM workflows since 2023. At a B2B marketplace with 15M supplier offers, I built a matching pipeline that halved the offers hidden from customers at over 99% precision.
As an AI engineer, I build LLM applications with evaluation, guardrails, and human-in-the-loop controls. I take projects end-to-end, from the business cost of errors to deployment and monitoring in production.
Work Experience
Senior Machine Learning Engineer
Maksmart
Aug 2024 - Apr 2026
Production ML and LLM systems for a B2B marketplace: 3M product cards, 15M supplier offers, 300K offer updates a day. Shipped 8 models to production across matching, catalog quality, pricing, and forecasting.
- Halved the number of supplier offers hidden from customers, matching them to product cards at precision above 99% with fastText retrieval and LightGBM reranking.
- Cut non-obvious duplicate product cards by 60% at 95% precision, merging them with an attention-based PyTorch scorer and graph-based grouping.
- Assigned product types to several million product cards using metric learning and LLM + RAG labeling, enabling grouping beyond existing catalog categories.
- Scored every incoming supplier offer for price anomalies so moderators reviewed the riskiest first, using a LightGBM ensemble with uncertainty estimates.
- Replaced fixed delivery estimates with data-driven targets for carriers, including routes without shipment history, while reducing a combined score of error and bias by 28%.
Core technologies: Python, PyTorch, LightGBM, scikit-learn, fastText, Elasticsearch, LLM + RAG, FastAPI, Docker, Kubernetes, PostgreSQL, ClickHouse.
Data Scientist
Architech NYC
Apr 2022 - Jul 2024
Only developer on each ML and LLM project for a container transportation company, working with the owner and dispatchers from requirements to production: rate calculation, email handling, document checks, and reporting.
- Built a rate suggestion service that dispatchers used for customer quotes, at 11% RMSPE on future periods, with daily retraining gated by validation.
- Sped up email replies with an LLM assistant that pulled shipment data and calculator rates, sending status updates automatically and drafting quotes for review.
- Automated driver document validation with OCR, signature detection, and cross-checks against internal shipment records, sending uncertain cases to dispatcher review.
- Let the owner track operations with Apache Superset dashboards, from late drivers and container yard time to order volumes and sales rep performance.
Machine Learning Engineer
Sape
Dec 2020 - Mar 2022
ML for SEO, search, and ad exchange products, with three models shipped to production.
- Replaced tag-based and random backlink discovery with embedding-based semantic search that got positive user feedback and further company investment.
- Shipped the search over 48M pages as a Docker service with an HTTP API that the backlink marketplace team deployed and ran in their product.
- Assigned topics to publisher sites automatically across 68 categories, using fastText embeddings and logistic regression served through FastAPI.
- Replaced a model that recommended the minimum price for 55% of publisher sites with a combined site score and a price curve fitted on past deals.
Ruby on Rails Developer
Sape
Dec 2016 - Nov 2020
Backend for a self-service ad platform running campaigns across Facebook, Google Ads, VK, and myTarget.
- Kept campaign statuses and spend in sync with each ad network's API so that client balances were charged correctly.
- Unified the API for external users and the frontend, removing duplicated integration logic.
- Moved CI and hosting from a compromised Jenkins server to self-hosted GitLab with configuration in the repository.
Earlier Experience
- Ruby on Rails Developer, ORDIN, Apr 2016 - Nov 2016: Rails apps for real estate and qualification workflows.
- Software Engineer, Nasca LTD, Dec 2012 - Sep 2013 and Apr 2015 - Jan 2016: C++ trading systems.
Core Skills
Machine Learning:
Python,
PyTorch,
LightGBM,
scikit-learn,
gradient boosting,
NLP,
embeddings,
fastText,
metric learning,
active learning,
anomaly detection,
forecasting,
uncertainty estimation,
model evaluation
Search, Matching & Ranking:
information retrieval,
semantic search,
vector search,
candidate generation,
reranking,
product matching,
duplicate detection,
Elasticsearch,
FAISS
LLM & RAG:
Large Language Models (LLMs),
Retrieval-Augmented Generation (RAG),
agent workflows,
tool calling,
MCP,
structured output,
prompt engineering,
LLM evaluation,
guardrails,
human-in-the-loop,
OpenAI API,
Claude Code
Production & Engineering:
FastAPI,
Docker,
Kubernetes,
Google Cloud Platform (GCP),
CI/CD,
model deployment,
model monitoring,
scheduled retraining,
Airflow,
PostgreSQL,
ClickHouse,
SQL,
Pandas,
Sentry
Education
Pskov State University
Master's degree,
Information Systems and Technologies
2008 - 2013
Languages
English
(Professional working proficiency),
Russian
(Native)