CV
Contact Information
| Name | Michael Diskin |
| Professional Title | Head of LLM R&D |
| [email protected] |
Professional Summary
AI/ML leader with a track record of bridging top-tier research and large-scale production. Built and scaled LLM and embeddings organizations from the ground up (30+ people, 4–5 teams).
Experience
-
2024 - present Moscow, Russia
Head of LLM R&D
Wildberries
- Built LLM R&D org from 0 to 30+ engineers (4–5 teams); manage a 500+ GPU cluster and define platform strategy for LLM and embeddings company-wide.
- Shipped universal text embeddings (retrieval, ranking, classification) and marketplace-scale MT translating tens of millions of product listings into 10+ languages, including low-resource.
- Deployed optimized LLM serving with use-case routing, batching, and quantization — cut GPU costs by 40%+ while meeting latency SLAs under peak traffic.
- Built RAG-powered assistants (internal and seller-facing) with safety guardrails, evaluation framework, and gold-standard regression datasets.
- Established research-to-production operating model: experiment standards, evaluation gates, reliable releases — reduced iteration cycles from weeks to days.
-
2022 - 2023 Tbilisi, Georgia
Senior Research Engineer
Brask AI
- Led R&D on a lip-sync model for AI video dubbing: research, model iteration, production integration.
- Improved robustness across diverse speakers and conditions; defined quality metrics and failure-analysis pipeline with product and engineering.
-
2021 - 2022 Moscow, Russia
Research Scientist
Yandex Research
- Research on efficient and distributed training for large models; 5 papers at NeurIPS, ICML, and ICLR.
- Led pre-training of a fully open-source Russian BERT-class language model; co-authored the Hivemind decentralized training library.
- Created an evaluation benchmark for graph neural networks under heterophily (460+ citations, ICLR 2023).
-
2020 - 2021 Moscow, Russia
Research Engineer
Huawei
- Computer vision research: optical flow and depth estimation for image-based localization.
-
2019 - 2020 Moscow, Russia
ML / Software Engineer
Early-stage startups
- Built ML-powered backend services and data pipelines from scratch in small teams; end-to-end ownership from prototyping to deployment.
-
2017 - 2018 Moscow, Russia
Software Engineering Intern
Yandex
- Large-scale analytics over multi-terabyte log data using internal MapReduce infrastructure (YT).
Education
-
2022 - 2024 Moscow, Russia
-
2019 - 2021 -
2014 - 2019 Moscow, Russia
Skills
ML & AI (Expert): LLM pre-training & alignment (SFT, DPO, RLHF), parameter-efficient tuning (LoRA, QLoRA), text embeddings & rerankers, seq2seq / machine translation, RAG pipelines, multimodal vision-language models
Frameworks (Expert): PyTorch, Hugging Face (Transformers, PEFT, TRL, Datasets), DeepSpeed, FSDP, Megatron-LM, vLLM, TGI, Triton Inference Server, TensorRT-LLM, ONNX Runtime
Data & Eval (Expert): COMET, BLEU, MTEB, LLM-as-judge, custom evaluation harnesses, human annotation pipelines, A/B testing
Infra & MLOps (Expert): CUDA, Docker, Kubernetes, Airflow, Weights & Biases, MLflow, Grafana / Prometheus, S3 / MinIO, Git, CI/CD
Languages (Expert): Python, C++, Bash, SQL
Awards
-
2024 HSE FCS Scholarship for Research Excellence
HSE University
-
2023 Tinkoff Education Scholarship
Tinkoff
-
2020 Xeek.ai "Put it on the Map!" — 2nd place (100+ teams)
Xeek.ai
Cash prize
-
2019 Kaggle "Recursion Cellular Image Classification" — 13th of 800+ teams
Kaggle
-
2019 Huawei Image Inpainting Hackathon — 2nd place (100+ teams)
Huawei
Cash prize
-
2019 International Data Analysis Olympiad — 16th of 1000+ participants
IDAO
Languages
Russian : Native speaker
English : Fluent