Looking for a job

Aleksandr Shchegolev

@aleksandrshchegolev

Message

I build production AI agents, RAG systems, and ML products for fintech.

Serbia

Message

What I'm looking for

I'm excited by opportunities to build end-to-end AI and ML products, explore new growth areas, and contribute to a strong team while growing as a professional.

I've built production AI systems for ВБ-digital and Gazprombank, from voice agents for collection calls to RAG-powered support automation.

At ВБ-digital, I developed a voice agent that automated 70% of soft collection, with capacity equivalent to 25–30 employees. I also built hierarchical LLM classification, document-processing, and Airflow workflows for operational analytics.

At Gazprombank, I improved a technical-support chatbot's operator transfer rate from 35% to 18% and reduced Customer Effort Score by 17%. I also developed debtor call automation and graph-based anti-fraud systems that increased detected real fraud by 13.5%.

Earlier at beeline Russia, I deployed CatBoost models for default and churn prediction, improving churn accuracy from 68% to 82%. I work across Python, LLMs, RAG, NLP, ML infrastructure, data pipelines, and end-to-end product delivery.

Experience

Work history, roles, and key accomplishments

Current

Artificial Intelligence Engineer

Current

ВБ-digital

Oct 2025 - Present (1 year)

Designed and developed the agentic module Developed a voice agent for collection calls.
The agent is powered by a RAG module that helps the model with the next answer based on the historical data from the operators.
- Automated 70% of soft collection with the agent.
- 1.5s for 95p end-of-speech -> beginning of audio
- Automation results: 60% of answered calls passed verification, 55% of authorized

Data Scientist

Gazprombank

Jan 2023 - Oct 2025 (2 years 9 months)

Developed a chatbot for the bank’s technical support.
• Built a RAG system based on Llama 2 model and LangChain framework. Used Milvus vector database to store knowledge
base (regulations, FAQ) for subsequent retrieval.
Implemented an Airflow pipeline for periodic data reloading into the database and model monitoring in Grafana.
• Implementation results: operator transfer rate decreased from 35% t

Data Scientist (Classic ML)

beeline Russia

Feb 2022 - Dec 2022 (10 months)

Predicted rates of customer defaults with gradient boosting-based models..
• Conducted EDA on customer data, tested models on open datasets to evaluate precision and recall, then trained on customer
data.
• Trained a CatBoost-based scoring model, which increased ROC-AUC by 4.2% and decreased inference time by 170ms
• Deployed the model in production.
Predicted client churn.
• Predicted which custo