Irina Buntina

Data Scientist

Klaviyo

• Developed a CLV prediction model for e-commerce customers that enabled high-value audience segmentation and increased revenue per recipient by 3–5% in pilot email/SMS campaigns. • Built a churn risk prediction model based on purchase history and email engagement, increasing repeat purchase rate by 0.5–0.9 percentage points in retention campaigns. • Implemented a send-time optimization model that selected the best email delivery window by segment and improved open rate by 2–4% without increasing unsubscribe or spam metrics.

NLP Engineer

TransTeleCom

• Developed an LLM-based sales script generation service for customer segments, products, funnel stages, and typical objections, increasing first-contact conversion by 8.4%. • Fine-tuned Qwen2-7B-Instruct with QLoRA on internal sales scenarios to improve a RAG-based sales script generator reducing major edit rate from 30–35% to 18–23%. • Fine-tuned a BERT-like encoder for multi-label text classification of ASR-based support dialogues, enabling automatic QA checklist scoring and cutting initial review time from 8–10 to 4–5 minutes per case.

Senior NLP Engineer

Sber

• Developed a GigaChat-based Text-to-SQL AI agent that reduced ad-hoc SQL requests to data engineers by 32.4% and saved around 80 person-hours per month for the B2C reporting team. • Built an AI-agent application for BI analysts that automatically generated and uploaded summaries of B2C master monitoring dashboards, reducing management reporting preparation time by 3 hours per week. • Developed an AI-powered document routing agent for SberDocs that reduced the average document approval cycle by 25–30% and saved headquarters secretaries 30–40 minutes per day.

Bachelor of the Faculty of Physics (direction of Physical Informatics)

NSU

Master’s degree in Information Management and Processing Systems

HSE