Senior Applied ML / GenAI Engineer

Shanghai · Open to worldwide remote work and relocation
Email: nikolai.zakharov92@gmail.com · LinkedIn · GitHub

English CV (PDF) →  |  中文简历 PDF →  |  中文简历 →

Summary

Senior Applied ML / GenAI Engineer with 8+ years spanning academic ML research and industry. Delivered LLM, multimodal, autonomous-driving, speech, and 3D generative systems across model training, evaluation, optimized inference, and infrastructure; Tsinghua M.S.

Professional Experience

AI Startup — ML Engineer

Generative Models & ML Platform · May 2026 – Jul 2026 · Shanghai

  • Worked on dental-crown generation models and built supporting multi-GPU/Kubernetes infrastructure.

VK — ML Engineer / Applied Research

Social media platform · Jan–Oct 2025; Feb–Apr 2026 · Moscow

  • Built LLM captioning and three-stage distillation for music-video recommendation; deployed an item2item quality dashboard for team evaluation.
  • Built TV-series detection and metadata extraction from regex-bootstrapped, LLM-generated data; lifted TVT by 4.5–6.3% and used content-model signals to filter same-episode uploads, cutting duplicates from 60% to 7%.

Kandinsky Lab (Sber AI) — Senior ML Engineer

GenAI · Nov 2025 – Jan 2026 · Moscow

  • Fine-tuned Qwen with LoRA for prompt enhancement; built multi-LLM synthetic data and reproducible multi-GPU image-to-video evaluation on A100s.

NIO — Algorithm Engineer, Autonomous Driving

Smart EV and autonomous driving · Jul 2023 – Jan 2025 · Shanghai

  • Trained an end-to-end transformer planner over map/visual context with distributed training and Hydra experiments.
  • Built C++ rule-based event tagging for open/closed-loop evaluation and automatic collection of failure cases.

T-Bank — ML Engineer, Voice Kit

Digital bank · Sep 2021 – Feb 2023 · Moscow

  • Trained/deployed voice conversion (95% speaker similarity) for a 10M+ MAU assistant; built Go streaming/tensor components and TensorRT edge/cloud inference.

Huawei — ML Engineer, CBG

Global ICT company · Jul 2019 – Sep 2021 · Beijing

  • Built auto-updating trie NLU (90%+ traffic); shipped on-device voice cloning and contributed to its on-device training framework with Huawei Russia; prototyped heterogeneous on-device RNN-T ASR; wrote ARM NEON gaze kernels slightly faster than OpenCV on ARM.

Education and Research

Tsinghua University · Beijing

  • M.S. Computer Science (Machine Learning), full scholarship; one year of PhD coursework at Tsinghua (2022–2023).
  • Huawei Noah’s Ark Lab, Research Intern (2018–2019): reimplemented DARTS in TensorFlow 1 with static graphs runnable on devices via TFLite; researched GAN-based face-recognition augmentation.
  • JVCIR: Towards Controllable Image Descriptions with Semi-Supervised VAE.

Technical Expertise

  • Models: PyTorch, transformers, LLM fine-tuning and distillation, 3D generative modeling, synthetic data, distributed and multi-GPU training
  • Evaluation: Multimodal, geometric, and open/closed-loop evaluation, dataset versioning, experiment management
  • Engineering: Python, C++, Go, TensorRT, TFLite, ARM NEON, model serving, Kubernetes/Kubeflow, Kueue, Docker, CI/CD
  • Languages: Russian (native); English and Mandarin Chinese (fluent); French and Turkish (basic)