source.tex

% basic part begin
\documentclass[10pt,a4paper]{report}
\usepackage[utf8]{inputenc}
\usepackage[russian]{babel}
\usepackage[OT1]{fontenc}
\usepackage{amsmath}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage[misc]{ifsym}
\usepackage{makeidx}
\usepackage{wasysym}
\usepackage{enumitem}
\usepackage{csquotes}
\usepackage{microtype}
\usepackage{vwcol}
\usepackage{tgbonum}
\usepackage[dvipsnames]{xcolor}


% basic part end

\usepackage{multicol}
\usepackage{fontawesome}
\usepackage{hyperref}
\usepackage[left=1cm,right=1cm,top=1cm,bottom=1cm]{geometry}

\setromanfont{LeagueSpartan}[
    Path=./Font/,
    Extension = .ttf,
    UprightFont=*-Regular,
    BoldFont=*-SemiBold
    ]

\author{Vladimir Baikalov}

\setlength\columnsep{10mm}
\setlength\parindent{0pt}
\pagenumbering{gobble}

\begin{document}

\noindent
\begin{minipage}[c]{0.66\linewidth}
\Huge{\textbf{Vladimir Baikalov}}

\small{R\&D Research Engineer / PhD Student}
\end{minipage}
\begin{minipage}[c]{0.33\linewidth}
\Letter: nonameuntitled159@gmail.com

\faLinkedin : \href{https://www.linkedin.com/in/noname-untitled}{linkedin.com/in/noname-untitled}

\faGithub:  \href{http://www.github.com/NonameUntitled}{github.com/NonameUntitled}

\faSend:  \href{http://www.t.me/noname\_untitled}{t.me/noname\_untitled}

\end{minipage}

\par\hbox{\Large{\textcolor{Plum}{\textbf{Work experience}}}}\kern3pt\hrule\kern5pt

\textbf{\large{Team lead R\&D RecSys Engineer/}}\textcolor{Plum}{Yandex Technologies}
\hfill
\textbf {\underline{June 2024 - Present}}

\textbf{Relevant areas}: RecSys, Information Retrieval, Highload, Deep Learning

{\leftskip=3mm
    \vspace{2mm}

    \textbf{Leading a team of three engineers}. Researching the application of e2e models instead of multi-stage classic recommender system pipeline. Investigating limitations and possibilities of applying semantic IDs for generating retrieval for stream music domain.

    \vspace{2mm}
    
    Deployed retrieval and ranking models for Yandex.Lavka (e-commerce service). This model increased \textbf{GMV by 2.1\%}. 
    
    \vspace{2mm}
    
    Publications: 
    
    \hspace{3mm} Short paper \textbf{\enquote{\href{https://dl.acm.org/doi/abs/10.1145/3705328.3748033}{Correcting the LogQ Correction: Revisiting Sampled Softmax for Large-Scale Retrieval}}} -- Improving the LogQ correction presented at \textbf{Google, Deepmind}. Paper was accepted at RecSys'25
    
    \hspace{3mm} Full paper \textbf{\enquote{\href{https://arxiv.org/abs/2507.15994}{Scaling Recommender Transformers to One Billion Parameters}}} -- demonstrating scaling laws in recommender systems and presenting novel pretraining idea. 
    
    \hspace{3mm} Reproducibility paper \textbf{\enquote{\href{https://dl.acm.org/doi/abs/10.1145/3705328.3748163}{Yambda-5B — A Large-Scale Multi-Modal Dataset for Ranking and Retrieval}}}. The largest dataset of streaming music data. Paper was accepted at RecSys'25.
    
    \vspace{2mm}
    
    Competitions: 
    
    \hspace{3mm} With the team took the 4th place at RecSys'25 challenge. The solution \textbf{\enquote{\href{https://dl.acm.org/doi/full/10.1145/3758126.3758131}{Blending Sequential Embeddings, Graphs, and Engineered Features}}}. Paper was accepted at RecSys'25.
    
    \vspace{2mm}
}

\textbf{\large{Machine Learning Engineer/}}\textcolor{Plum}{Google, YouTube}
\hfill
\textbf {\underline{January 2023 - June 2024}}

\textbf{Relevant areas}: Transformers, Highload, Large Language Models (LLMs)

{\leftskip=3mm
    \vspace{2mm}
    
    \hspace{3mm} Developed video games detection algorithm for shorts. The final model achieved \textbf{4\% recall boost without precision drop} and allowed to \textbf{conduct fine-tuning more frequently}. \textbf{Designed an end-to-end system} for serving this classifier, the solution is now being hosted by YouTube Gaming team.
    
    \vspace{2mm}
    
    \hspace{3mm} Worked on deployment of LLMs in production including improvement of \textbf{fine-tuning and distillation} pipelines.
    
    \vspace{2mm}
}
\textbf{\large{Deep Learning Engineer/}}\textcolor{Plum}{Yandex Technologies}
\hfill
\textbf {\underline{August 2021 - December 2022}}

\textbf{Relevant areas}: RecSys, Information Retrieval, Highload, Deep Learning


{\leftskip=3mm
   
    \vspace{2mm}
    
    \hspace{3mm} Transferred an ML model from experimental setup (Python, PyTorch) to production framework (C++, YNMT).  Supported weekly continuous fine-tuning process for \textbf{Yandex.Ads}. \textbf{It is still being applied in production.} This model increased \textbf{GMV by up to 1.5\%} and \textbf{CTR by up to 5\%}. The result is verified by AB tests and experiments.
    
    \vspace{2mm}
        
    \hspace{3mm} Applied encoder-based models for improving personalized ads and search recommendations. The current solution \textbf{boosts production metrics by up to 6\%}.
    
    \vspace{2mm}
        
    \hspace{3mm} Implemented multiprocessing Python package (YtReader) for fast and convenient data preprocessing. The final solution reduced the time required for model training/evaluation \textbf{by up to 5 times}.
    
    \vspace{2mm}
}

\textbf{\large{Machine Learning Engineer}}/\textcolor{Plum}{Huawei R\&D Dept.}
\hfill
\textbf {\underline{March 2020 - July 2021}}

\textbf{Relevant areas}: Object Detection/Tracking, Digital Sound Processing, Optical Character Recognition.

{\leftskip=3mm
    \vspace{2mm}
    
    \hspace{3mm} Algorithm for vehicles trajectories prediction using radar data only. This approach now is being used in a real-world application. Proposed solution \textbf{outperforms the previous algorithm by 44\%}. 
    
    \vspace{2mm}
    
    \hspace{3mm} Algorithm for the knuckle-knock sound pattern detection. Proposed architecture achieved \textbf{90\%} Precision and~\textbf{95\%} Recall. With the usage of touch sensors precision was improved up to \textbf{94\%}.

    \vspace{2mm}
}

\par\hbox{
\Large{\textcolor{Plum}{\textbf{Education}}}
}{\kern5pt\hrule\kern5pt}

\textbf{\large{PhD student/}}\textcolor{Plum}{ITMO University, RecSys dept.}
\hfill
\textbf{\underline{September 2024 - Present}}

\vspace{2mm}

\hspace{3mm} Working on improving retrieval models by applying enhanced user/item representations.

\vspace{2mm}

\textbf{\large{Master degree/}}\textcolor{Plum}{Skoltech, DS major}
\hfill
\textbf{\underline{September 2021 - July 2024}}

\vspace{2mm}

\hspace{3mm} \textbf{GPA: 4.5 / 5.0}. Thesis project: \href{https://arxiv.org/abs/2403.00895}{End-to-end Graph-Sequential Representation Learning for Accurate Recommendations}.

\vspace{2mm}

\hspace{3mm} Publications: Short paper was accepted at \textbf{TheWebConf'24 conference (A*-tier)}, Singapore. 

\vspace{2mm}

\hspace{3mm} Received certificate of achievement \textbf{"Best Research Thesis"}.

\vspace{2mm}

\textbf{\large{Bachelor degree}}\textcolor{Plum}{/ITMO University, CS major}
\hfill
\textbf{\underline{September 2017 - August 2021}}

\vspace{2mm}

\hspace{3mm} \textbf{GPA: 4.7 / 5.0}. Developed multi-agent policy-based algorithm \textbf{REM (Reinforce, Embedding, Monte-Carlo)} for baggage handling system.

\vspace{2mm}

\hspace{3mm} Publications: \textbf{\href{https://aaltodoc.aalto.fi/handle/123456789/111642}{\enquote{Multi-Agent Deep Reinforcement Learning-Based Algorithm For Fast Generalization On Routing Problems}}} was published at the \textbf{YSC 2021 conference}. This project was done in collaboration with \textbf{Aalto University, Finland}.

\vspace{2mm}

\end{document}