Lead ML Engineer | LLM Inference

Remote

Looking for not just an ML Lead, but a very strong hands-on engineer who has actually worked with large LLM inference in production.

Scale is crucial here: models up to 1T parameters, distributed inference, GPU clusters, and tasks where you need to simultaneously improve latency, throughput, cost, and stability.

What the job involves: • production inference of large LLMs • distributed inference and serving optimization • SGLang - already used in production • post-training LLM • agent harness • caching, batching, quantization, speculative decoding • working with GPU infrastructure - currently 32×H200 + AWS, expansion planned • technical leadership of the NLP/CV direction

The role is approximately 70% hands-on. No people management, performance reviews, or 1:1s here.

What is critical: • strong production LLM inference • experience with distributed inference of large models • vLLM / SGLang / TensorRT-LLM or comparable stack • understanding of tensor/pipeline/expert parallelism, KV cache, GPU optimization • post-training: SFT / DPO / RLHF, etc. • strong foundation in ML / CS / mathematics

A big plus: CUDA/Triton, DeepSpeed, Megatron-LM, FSDP, MoE, own publications, PhD, research, or significant open-source contributions.

We are not looking for someone who has primarily worked with RAG, LangChain, API integrations, or simply deployed ready-made LLMs. Deep work with the models themselves and their inference is required here.

🌍 Remote, B2B (outside Russia, Belarus) 🕐 CET-friendly timezone 🔞 The product contains 18+ content

If you have personally accelerated large LLMs in production and understand what happens with the model at the GPU, serving, and distributed systems level - write to me in private messages.

@SelitskayaDaria (location, salary expectations, summary of requirements)