Nikita Barinov
Available for opportunities

Nikita Barinov

Machine Learning Engineer

Building intelligent systems with LLMs, RAG, and multimodal AI. Passionate about pushing the boundaries of what's possible with machine learning.

Get in Touch +7 (905) 133-08-50 Telegram GitHub
3+
Years Experience
10+
Projects
4
Companies

About Me

Passionate about AI and its real-world applications

I am a Master's graduate of MIPT and currently work as a Machine Learning Engineer at a leading tech company. I have experience across multiple Big Tech companies, where I worked on both research-driven tasks and applied business projects. My background includes developing and deploying machine learning solutions in production, collaborating with cross-functional teams, and translating research ideas into practical systems.


At the moment, my main interests lie in Retrieval-Augmented Generation (RAG) systems and machine learning for audio. I enjoy working at the intersection of research and real-world applications, quickly learning new approaches, and tackling complex problems with a pragmatic mindset.

Experience

My professional journey in ML

ML Engineer @ AI VK
Sep 2025 – Present
Python YQL LLM RAG VLM PyTorch MLflow Docker Triton Inference Server
  • Fine-tuned LLM models on internal datasets to improve embedding space quality.
  • Ran 100+ experiments with different training setups, increasing dataset diversity +20%.
  • Developed multimodal content models for VK Video, improved recall by 14% while maintaining precision at 0.9.
  • Trained relevance VLM and improved recall by +30% compared to previous model.
  • Deployed models using Triton Inference Server.
ML Engineer @ Tinkoff AI Center
Sep 2024 – Sep 2025
Python SQL RecSys Transformers Sequential Models Airflow Docker Linux ML Pipelines CatBoost Experiment Tracking
  • Trained Transformer-based sequential recommender models for candidate generation, timespent +7%.
  • Improved recall of candidate generation stage by introducing sequence-aware embeddings.
  • Maintained production ML pipelines: batch inference, retraining, monitoring, data validation.
  • Automated training and deployment using Airflow + Docker.
  • Worked with large-scale logs using SQL and internal data storage.
Data Scientist @ Sber AI
Feb 2024 – Aug 2024
Python PyTorch Diffusion Stable Diffusion Grounded-SAM LLaMA LLaVA Computer Vision WebDataset Accelerate Distributed Linux
  • Built large-scale instruct image-editing dataset generation pipeline using diffusion models + VLM + LLM.
  • Generated millions of samples using Stable Diffusion + Grounded-SAM + LLaMA + LLaVA-NeXT.
  • Implemented distributed inference pipeline with HuggingFace Accelerate.
  • Optimized storage and loading using WebDataset to support high-throughput training.
  • Accelerated generation pipeline using OneDiff and batching strategies.
ML Engineer Intern @ Yandex Music
Jun 2023 – Nov 2023
Python SQL/YQL RecSys Transformers Embeddings CatBoost Feature Engineering Pandas NumPy Linux
  • Integrated Transformer-based track embeddings into next-track prediction pipeline used in production recommendation system.
  • Improved candidate ranking quality by enriching feature space with sequence-aware embeddings.
  • Designed feature generation pipeline in Python + YQL for large-scale logs.
  • Validated improvements using offline metrics and A/B-ready evaluation setup.

Skills & Stack

Technologies and tools I work with

LLM & RAG

LLM LangChain RAG Agentic RAG GraphRAG Qdrant Vector Search Fine-tuning

Deep Learning

PyTorch Transformers Multimodal Models Diffusion Models Audio ML Whisper

ML in Production

Python SQL YQL MLflow Triton Inference Server

Recommendation Systems

Sequential Models Transformers Vector Search CatBoost

Computer Vision

Stable Diffusion Grounded-SAM LLaVA Object Detection

Infrastructure & MLOps

Docker Kubernetes FastAPI Airflow PostgreSQL Distributed Training WebDataset

Projects

What I've been building

Audio RAG Apr 2026 – Present
Python Whisper Triton RAG Qdrant BGE-M3 LLM Docker
  • Built production-ready RAG system for audio content with Triton Inference Server.
  • Implemented speech-to-text with Whisper ASR and word-level timestamps.
  • Integrated BGE-M3 multilingual embeddings with Qdrant vector database.
  • Deployed Docker-based infrastructure with scalable inference serving.
View on GitHub
RAG Architectures Comparison Sep 2025 – Present
Python RAG LangChain GraphRAG Agentic RAG LLM API Kubernetes
  • Implemented vanilla, graph-based and agentic RAG pipelines.
  • Built evaluation framework for multi-hop QA.
  • Used local LLM + embedding server.
  • Compared retrieval quality across different architectures.
View on GitHub
Audio Question Answering Jan 2026 – Feb 2026
PyTorch torchaudio Transformers Whisper Qwen Adapter Audio
  • Implemented AudioLLM architecture with Whisper encoder + LLM.
  • Built adapter-based training pipeline.
  • Generated QA dataset from LibriSpeech.
  • Trained model using HuggingFace Trainer.
View on GitHub
AI News Meme Generator Jun 2025 – Jul 2025
Python LangChain Ollama RAG Streamlit PostgreSQL FAISS
  • Built automated news scraper for Russian news sources (Lenta.ru, RIA.ru).
  • Implemented topic classification using LLaMA 3.1 with few-shot learning.
  • Created RAG pipeline for context-aware joke generation.
  • Developed web interface with semantic search and real-time meme generation.
View on GitHub
ShopSpy — Price Tracker Mar 2026 – Present
Python JavaScript FastAPI SQLite Chrome Extension Telegram Bot LLM API
  • Built browser extension for price tracking on Wildberries and Ozon.
  • Implemented fake discount detector based on price history analysis.
  • Integrated AI-powered review summary using Gemini API.
  • Created Telegram bot for real-time price drop notifications.
View on GitHub

Publications

Research contributions

Garage: Generative Augmentation Framework for Transforming Object Representations in Images

2024

Python framework for automatic object replacement in images. Uses GroundingDINO for object detection and PowerPaint for inpainting to generate AI-based alternatives. Supports large-scale dataset augmentation with Gradio interface.

Python PyTorch Diffusion PowerPaint GroundingDINO Gradio
PDF

Education

My academic background

Moscow Institute of Physics and Technology (MIPT)

Sep 2020 – Present

Applied Mathematics and Computer Science

Department of Data Analysis, MIPT & Yandex School of Data Analysis

Sep 2022 – May 2024

Specialization: Data Analysis (Yandex School of Data Analysis)

Lyceum №2, Rybinsk

Sep 2009 – Jun 2020

Physics and Mathematics class

Courses & Certifications

Continuous learning

YSDA Speech Processing Course

Yandex School of Data Analysis
Sep 2025 – Jan 2026
Python Deep Learning VAD SED CTC Transfer Learning RNN-T ECAPA-TDNN ASR
  • Implemented and trained different DSP models, such as ECAPA-TDNN, RNN-Transducer, finetuned ASR models.

AI Agents Intensive: From LLM Queries to Agent Systems

Yandex
Apr 2026
LLM AI Agents MCP Multi-Agent Systems Guardrails Memory Agent Evaluation Production Engineering
  • Studied modern AI agent architecture: thought-action-observation cycle, tools, memory, and multi-agent orchestration
  • Learned production engineering principles: monitoring, scaling, and deploying agent systems under real load
  • Download Certificate