Aleksei Stepin Email & Phone Number

• Built an NLP-driven system to match MOOCs to a skills taxonomy using vector search and LLMs. The system performed a two-stage process based on vectors and LLMs.• Tackled a low-resource language NER challenge by developing a robust annotation pipeline, overseeing annotators, curating a comprehensive dataset. Achieved high model quality (F1 score > 0.9) while maintaining processing speed.• Developed classification systems for low-resource languages, leveraging a range of models from simple Tf-Idf+SGD to sophisticated LLMs, tailored to task complexity and optimizing the quality-speed tradeoff.• Created and deployed a cloud-based captcha recognition service with 0.98 accuracy rate, handling 300+ qps.

• Designed and developed a product for avatar generation based on Stable Diffusion and accessible as a Telegram bot• Deployed it to AWS and scales for 100+ user’s requests at the same time• Divided the monolith into independent services, one of which is the core of the generative services ecosystem• Together with an intern, conducted 50+ experiments with SoTA generative networks. This formed the basis of the new product• Designed WebSite architecture• Recruited and onboarded 3 developers, now leading website development.

Москва, Москва, Россия

• Maintained and improved high-loaded ML pipelines with millions of QPS• Designed and implemented product experiments to improve advertising ranking• Made A/B tests of my experiments, both with pipelines and product improvements• Contributed in Logos, an in-house analog of Airflow for managing big-data pipelines• Prepared and gave some lectures for a team( 10 members), e.g. a lecture about AB-systems: at all and about our one

Москва, Россия

• Created the infrastructure for the DS team (10 people): worked out a plan for rolling out projects to production and infrastructurefor this task, ordered the necessary equipment, and set it up.• Enabled data-driven decision-making for businesses through project profitability evaluations and building payback equations• Worked on the problem of product clustering• Created and delivered docker-course in 4 lectures, with presentations and homework for data-office ( 20 members)

Нижний Новгород, Нижегородская Область, Россия

• Created modules of big web systems. Designed and implemented modules, API, and DB schemas• Wrote a library for loading xml data from tablets into a database• Organized in-office Computer Science Club with lectures about different CS topics. Lecture• Stack: C#, SQL, JS, TS, React.

Master'S Degree, Artificial Intelligence (Ai)

Activities and Societies: Conducting research on Large Language Models (LLMs), with a particular focus on their safety and responsible.

Бакалавр, Information Technology