Looking for a job

Руслан User

@user744

Data Scientist (NLP / LLM) at Россельхозбанк who improved call-center resolution from 45% to 68% with an on-premise LLM assistant.

Russian Federation

Message

What I'm looking for

I’m looking for a full-time Data Scientist (NLP / LLM) role that can be done remotely.

At Россельхозбанк, I developed an LLM assistant for the corporate call center, deployed on the bank’s own infrastructure without external APIs. After a shadow rollout and A/B test, resolved rate rose from 45% to 65%, reaching 68% after full launch.

I designed a LangGraph multi-agent system and fine-tuned Qwen 2.5 32B with LoRA on 10,000 labeled conversations. I also built a RAG system over 20,000 pages of regulations, achieving Recall@5 of 0.86.

At Boxberry, I delivered an end-to-end customer-support LLM agent with RAG. Its A/B test increased resolved rate from 52% to 74% and reduced operator workload by 30%.

At Boxberry, I also worked on customer-message classification and an OCR-plus-NER pipeline for waybills. The classification model reached 0.89 macro F1, while the document pipeline achieved 86% Exact Match and reduced batch processing time from 30 to 5 minutes.

Experience

Work history, roles, and key accomplishments

Data Scientist (NLP / LLM)

Россельхозбанк

Oct 2024 - Apr 2026 (1 year 6 months)

Developed an LLM assistant for a corporate call center, deployed on the bank's own infrastructure. Achieved a resolved rate increase from 45% to 68%, reduced response time from 12 to 8.4 minutes, and improved customer satisfaction.

Python PyTorch Hugging Face Transformers PEFT LoRA LangGraph VLLM Qdrant BGE M3 Qwen2.5 Mistral fastAPI SQL Docker Git

Data Scientist (NLP / LLM)

Boxberry

Oct 2021 - Aug 2024 (2 years 10 months)

Led full-cycle ML projects for contact center and logistics, including an LLM agent with RAG, customer request classification, and OCR+N ER for invoice registration. Improved resolved rate from 52% to 74% and reduced operator load by 30%.

Python PyTorch Hugging Face PEFT LoRA LangGraph VLLM Qdrant Onnx OpenCV Tesseract DeepPavlov LabelStudio scikit learn fastAPI Docker Kafka SQL Git

Education

Degrees, certifications, and relevant coursework

Уфимский государственный нефтяной технический университет

Высшее, Автоматизация технологических процессов и производств

Окончил Уфимский государственный нефтяной технический университет в 2016 году по специальности «Автоматизация технологических процессов и производств».