Lead Machine Learning Engineer

For this vacancy, please call the number: 8 (800) 333-30-73. Responsibilities: Design and train large language and multimodal models (from 7B parameters): corpus preparation, tokenization, pretraining, SFT, RLHF/DPO. Organize distributed training on GPU clusters (FSDP, DeepSpeed ZeRO-3, Megatron-LM, tensor/pipeline parallelism), increase computational resource utilization. Develop and optimize custom CUDA/Triton kernels for training and inference. Compress models for industrial inference: quantization (INT4/FP8, GPTQ, AWQ), distillation, pruning, speculative decoding. Meet strict latency and cost requirements. Build RAG systems and industrial-scale agentic architectures with quality assessment, protection against hallucinations and prompt injection. Develop real-time recommendation and ranking systems at loads from 100,000 RPS. Design the company's ML platform: feature store, model registry, model CI/CD, automatic retraining, monitoring of data drift and quality degradation. Design and conduct online experiments: A/B tests, causal inference, uplift modeling. Conduct research at the state-of-the-art level, reproduce and adapt fresh scientific papers to business tasks. Be responsible for the architecture of the ML direction, conduct code review. Defend technical solutions before management, evaluate the financial effect of models.

Requirements

Higher education in mathematics, applied mathematics, computer science, or physics. At least 8 years of experience in commercial ML system development, of which at least 3 years in a lead or principal engineer role. Experience in training and fine-tuning large language models in a distributed environment on multi-node GPU clusters. Expert knowledge of Python and C++, experience writing CUDA kernels or Triton kernels. Deep knowledge of PyTorch, including its internal workings (autograd, compile, distributed primitives). Experience with inference frameworks: vLLM, TensorRT-LLM, Triton Inference Server, ONNX Runtime. Deep mathematical knowledge: linear algebra, probability theory, mathematical statistics, optimization methods, information theory. Experience building high-load systems: Kubernetes, Kafka, gRPC, distributed storage, profiling and performance optimization. Experience with MLOps tools: MLflow, Kubeflow, Airflow, DVC, Feast, Ray. Experience with big data: Spark, ClickHouse, volumes from hundreds of terabytes. Knowledge of information security principles for ML systems: adversarial attacks, data protection, federated learning, differential privacy. English at a level of fluent reading and writing of technical documentation and scientific papers. Will be an advantage: Advanced degree (PhD) in machine learning or a related field. Publications at leading conferences (NeurIPS, ICML, ICLR, ACL, CVPR, KDD). Experience in pretraining LLM from scratch on large clusters (from 256 GPUs). Contribution to open-source ML projects (maintainer or active contributor to major libraries). Prize-winning places in data analysis competitions (Kaggle Grandmaster / Master or equivalent). Experience speaking at professional conferences.

Higher education — Bachelor's degree; Experience: 6 years