Nikolai Zakharov (罗一阳)

Senior Machine Learning Engineer (Multimodal / LLM)

Moscow, Russia (Open to China/Singapore/Global)
Tel: +86 136 2105 3870
Email: n.d.zakharov@outlook.com
LinkedIn: https://www.linkedin.com/in/nikolai-zakharov-285841a7/
Website: https://yiyang92.github.io/

Download English CV (PDF) →  |  下载中文简历 (PDF) →

Summary

Senior Machine Learning Engineer (Multimodal / LLM). 8+ years production AI at scale. Specialized in Audio LLMs, multimodal systems, inference optimization (vLLM, TensorRT). Delivered: captioning service for music recommendations (VK), RAG agents (+20%, NIO), voice synthesis (-88% latency, Tinkoff). Tsinghua MSc, 10 years China tech ecosystem. Fluent in Russian, English, Chinese.

Professional Experience

VK AI, Applied Research Team (Russia) — Senior Machine Learning Engineer

Jan 2026 – Present

  • Built LLM-based caption generation service improving recommendation relevance for music videos; designed 3-stage pipeline with distillation to train metadata extractor and audio-text relevance predictor; deployed item2item recommendation quality dashboard as key metric for recsys team.

VK Video, VK (Russia) — Senior Machine Learning Engineer

Jan 2025 – Dec 2025

  • Architected NER + metadata extraction pipeline (LLM + rules) for recommendation system; improved candidate generation quality, increased TVT by 25%, reduced duplicate results from 60% to 7%.

NIO (Shanghai, China) — Senior Machine Learning Engineer

Mar 2023 – Jan 2025

  • Designed AI Agent workflow (Golden Idea Platform): RAG retrieval + LLM assessment (+20%); built RAG-based LLM assistant; deployed C++ real-time AD event detection with on-car evaluation.

Tinkoff (T‑Bank) (Moscow, Russia) — Machine Learning Engineer

Sep 2021 – Feb 2023

  • Trained and deployed Russian Voice Conversion model (95% speaker similarity) with TensorRT edge optimization; reduced HiFiGAN vocoder RTF by 88%.

Huawei Technologies (Shenzhen, China) — Machine Learning Engineer (CBG)

Jul 2019 – Sep 2021

  • Built multilingual trie-based NLU engine serving large-scale voice assistant traffic (90%+ coverage); developed voice cloning system for cloud and edge.

Research Experience

Sber AI / Kandinsky Lab (Russia) — Research Collaboration

Dec 2025 – Jan 2026

  • Built prompt optimization system for Kandinsky diffusion model; designed distributed LLM evaluation infrastructure with GPU cluster orchestration.

Huawei Technologies — Research Intern (Noah Ark Lab)

2018 – 2019

  • Developed AutoML (DARTS) and GAN-based identity-preserving face recognition augmentation.

Education

Tsinghua University, Beijing, China

  • M.S. in Computer Science (Machine Learning), 2016 – 2019
    • Full Scholarship
    • Thesis: Diversified Image Captioning with Deep Learning (Chinese)
  • Ph.D. in Computer Science, 2022 – 2023
    • Completed coursework; returned to industry

Technical Expertise

  • LLM Systems: RAG pipelines, agent frameworks, model distillation, inference optimization (vLLM, TensorRT, quantization)
  • Multimodal: audio-text models, music understanding, captioning, diffusion models
  • ML Infrastructure: distributed training (FSDP, DeepSpeed), Kubernetes, KServe, GPU clusters
  • Languages: Python (PyTorch, HuggingFace), C++ (real-time inference), Go

Publications

中文简历

个人简介

高级机器学习工程师（多模态/LLM），8年以上生产级AI经验。专注音频大模型、多模态系统、推理优化（vLLM、TensorRT）。交付：音乐推荐描述服务（VK）、RAG智能体（+20%，蔚来）、实时语音合成（-88%，Tinkoff）。清华硕士，十年中国科技生态。精通俄语、英语、中文。

工作经历

VK AI / 应用研究组 — 高级机器学习工程师

2026.01 – 至今

  • 构建LLM描述生成服务，提升音乐视频推荐相关性；设计三阶段管道（蒸馏训练元数据提取器、音频-文本相关性预测器）；部署item2item推荐质量仪表板作为推荐团队核心指标。

VK Video / 视频推荐组 — 高级机器学习工程师

2025.01 – 2025.12

  • 架构推荐系统的NER+元数据提取管道（LLM+规则），提升候选生成质量，TVT提升25%，重复结果率60%降至7%。

蔚来汽车 / 自动驾驶大模型组 — 高级机器学习工程师

2023.03 – 2025.01

  • 架构AI智能体系统（金点子平台）：RAG检索+LLM评估（+20%）；构建RAG-based LLM助手；部署C++实时AD事件检测，支持车载评估。

Tinkoff银行 / 语音技术组 — 机器学习工程师

2021.09 – 2023.02

  • 训练并部署语音转换系统（Transformer，95%相似度），服务1000万+月活；优化HiFiGAN声码器（GPU算子融合，RTF降低88%）；架构TensorRT边缘部署。

华为 / 小艺语音助手 — 机器学习工程师

2019.07 – 2021.09

  • 构建多语言Trie树NLU引擎，服务大规模语音助手流量（90%+覆盖）；开发云端与端侧语音克隆系统。

研究经历

Sber AI / Kandinsky Lab — 研究合作

2025.12 – 2026.01

  • 构建Kandinsky扩散模型的提示词优化系统；设计分布式LLM评估基础设施，支持GPU集群编排。

华为诺亚方舟实验室 — 研究实习生

2018 – 2019

  • 开发AutoML (DARTS)；基于GAN的identity-preserving人脸识别数据增强。

教育背景

清华大学，北京，中国

  • 硕士（2016 – 2019），计算机科学与技术（机器学习），全额奖学金
    论文：《基于深度学习的多样性图像描述生成》
  • 博士（2022 – 2023），课程完成，返回产业界

技术专长

  • LLM系统：RAG管道、智能体框架、模型蒸馏、推理优化（vLLM、TensorRT、量化）
  • 多模态：音频-文本模型、音乐理解、描述生成、扩散模型
  • ML基础设施：分布式训练（FSDP、DeepSpeed）、Kubernetes、KServe、GPU集群
  • 编程：Python（PyTorch、HuggingFace）、C++（实时推理）、Go

论文发表

  • Towards Controllable Image Descriptions with Semi-Supervised VAE (JVCI&R, 2019)