Open to opportunities

Maksim Pukin

@maksimpukin

Message

Machine Learning Engineer specializing in NLP, LLMs, RAG, and recommendation systems.

Russian Federation

Message

What I'm looking for

I seek impactful ML roles building NLP/LLM and recommendation systems in production, with focus on measurable business impact, robust serving, and opportunities to optimize inference and model performance.

I am a Machine Learning Engineer with five years of experience building production NLP models, LLM-based solutions, and recommendation systems. I focus on applying LLMs, RAG, and embedding retrieval to drive measurable business outcomes, such as CTR, DAU, and revenue growth.

At EPAM I designed on-prem multi-agent RAG pipelines, fine-tuned ~14B LLMs with KL-constrained PPO to increase contact rates by 13–17%, and optimized inference to speed generation 1.7–1.9×. Previously, at Wildberries and Yandex I delivered recommendation and BERT-based NLP improvements that raised feature coverage, CTR, and automated routing rates.

I deliver end-to-end solutions—from data engineering and model training (PyTorch, CatBoost) to serving and monitoring (Docker, Kubernetes, FastAPI, vLLM)—and I enjoy solving production-scale ML problems that improve user experience and business metrics.

Experience

Work history, roles, and key accomplishments

Current

Machine Learning Engineer

Current

EPAM Systems

Dec 2023 - Present (1 year 11 months)

Designed an on-prem multi-agent RAG pipeline and LoRA text-to-SQL adapter reducing median root cause analysis time from 3h to 45min; fine-tuned a ~14B LLM with KL-constrained PPO raising contact rate by 13–17% on a 6–9M daily push A/B rollout and optimized inference with speculative decoding to speed generation ~1.7–1.9×.

LangGraph RAG PPO VLLM Speculative Decoding Monitoring Docker Kubernetes

Machine Learning Engineer

Wildberries

Aug 2022 - Nov 2023 (1 year 3 months)

Added purchase-frequency and customer-pattern features to a recommendation widget, yielding a 0.21% CTR uplift on target categories and positive GMV impact; expanded candidate pool with embedding-based retrieval increasing DAU who ordered from recommendations by 0.3%.

Embeddings CTR Optimization Python PyTorch A B Testing Clickhouse

Machine Learning Engineer

Yandex

Jun 2020 - Jul 2022 (2 years 1 month)

Built BERT-based models to extract POI attributes from 10M+ reviews, increasing coverage from 15% to 35% and boosting POI card CTR by 1.6%; improved feedback classification (+13 pp precision, +8 pp recall) and automated ticket routing from 45% to 64%; trained CatBoost to deduplicate POIs, reducing priority-category duplicates from 4% to 3%.