ml engineer for AI companions

Joi AI is a platform for AI-lationships: personalized, emotionally intelligent connections between people and AI characters. It serves emotional and intimate needs and has tens of millions of conversations. Joi Lab is its open research arm, building open-source generative models, agent architectures, and training infrastructure.

• Speed up and scale production LLM inference using SGLang, KV and prefix caching, batching, quantization, and speculative decoding
• Run distributed inference for very large models across multi-GPU and multi-node setups
• Benchmark new GPU servers and hardware, bring them into production, and adapt serving code
• Lead NLP and CV teams technically by reviewing experiments, setting direction, and addressing wrong turns early
• Train and fine-tune language models for AI companions and improve agent harnesses and chat algorithms
• Track inference and post-training research and open-source work, and turn findings into the ML roadmap
• Collaborate with validation, content, and dataset preparation teams to design experiments and measure model quality

• Deep hands-on experience optimizing production LLM inference with SGLang, vLLM, or TensorRT-LLM
• Experience with distributed inference or training of large models, including MoE, tensor/expert/pipeline parallelism, and multi-node GPU clusters
• Strong understanding of KV cache, attention kernels, batching, quantization, and GPU profiling
• Experience training and fine-tuning LLMs, including post-training such as RLHF or DPO
• Proven technical leadership through reviews, mentoring, and technical decisions while continuing to write code
• Proficiency with PyTorch, transformers, and related libraries
• Advanced English or Russian
• Будет плюсом: experience at AI-focused startups or companies such as Character AI or OpenAI, backend engineering with Python, Go, or C#, scalable deployment systems, CUDA or Triton kernel development, a computer vision background or experience accelerating generative image or video models, multimodal LLMs, first-author papers or notable open-source contributions, a degree in CS, math, or physics (MSc or PhD)

• 28 Days of annual leave
• 7 Yearly wellness days; no sick notes needed
• Referral rewards up to $5000
• 50% Of training, conference, and global meetup costs covered
• Subsidized English classes through a corporate discount
• Up to $1,000 annually for medical fees or private insurance if not on the group plan
• Office equipment or $1000 every three years for a home office or co-working setup
• Gratitude bonuses redeemable for swag, massage vouchers, or team adventures