ml engineer for llm performance optimization

The team is building a platform for AI-lationships, emotionally intelligent connections between people and AI characters, using open-source models and real AI agency rather than ChatGPT wrappers.

• Own the performance and technical direction of LLMs
• Optimize 1T+ parameter models to run faster and more cheaply on distributed GPUs
• Provide technical leadership to NLP/CV engineers through expertise

• Deep hands-on experience optimizing LLM inference with SGLang, vLLM, or TensorRT-LLM
• Experience with distributed inference and training, including MoE, tensor/expert/pipeline parallelism, and multi-node clusters
• Experience with KV cache, attention kernels, batching, quantization, and GPU profiling
• Experience training and fine-tuning LLMs, including post-training methods such as RLHF and DPO
• Proven technical leadership while continuing to write code
• PyTorch and transformers
• Advanced English or Russian
• Будет плюсом: CUDA/Triton, CV background, multimodal LLMs, open-source contributions, MSc/PhD

• 28 Days of leave plus 7 wellness days
• 50% Training coverage
• Health support up to $1000/yr
• $1000 Every 3 years for workspace