AI Platform Engineer

As an AI Platform Engineer at Trezor, you own a dedicated multi-GPU inference server end-to-end: the hardware and OS underneath it, the model serving stack on top, and the engineers who use it every day. Trezor is building its own AI infrastructure so employees can work with large language models on hardware Trezor controls, with no data leaving the premises. This is a hands-on, broad role where you will benchmark newly released models one day and help developers integrate agents into their workflows the next.

What you'll do

• Collaborate with IT to own the full lifecycle of the GPU server: OS, storage, networking, and set up observability for GPU utilization, thermals, memory, request latency and throughput with alerting that catches problems before users do
• Build and maintain an inference stack using tools such as vLLM, SGLang, LiteLLM as a gateway, and Open WebUI as the front end
• Deploy, upgrade and tune open-weight models, and evaluate new model releases as they land by benchmarking them for quality, throughput and latency
• Own quotas, routing and cost/usage reporting across teams, and collaborate on optimizing cloud usage when closed models are needed
• Be the go-to person for developers integrating internal models into their tooling such as IDE assistants, agents, CI pipelines and internal apps
• Maintain API keys, endpoints and documentation; write internal guides to make onboarding self-service and run enablement workshops and office hours
• Design and implement queueing and prioritization systems, deciding who has priority and how long-term tasks run overnight

What you bring

• Coding experience and Infrastructure as Code skills using tools such as Ansible, Terraform, or Docker/Kubernetes to automate your work
• Linux systems administration knowledge: networking, storage, containers, systemd, and troubleshooting from the kernel up
• Genuine interest in the open-weight model ecosystem and knowledge of which models matter currently
• A service mindset where you enjoy unblocking other engineers and documenting your work
• English for daily work

Nice to have

• Experience serving LLMs in production or a serious homelab using tools such as vLLM, SGLang, llama.cpp or Ollama
• Working knowledge of NVIDIA GPU operations: drivers, CUDA, NVLink, MIG, nvidia-smi and DCGM
• Datacenter experience: rack power budgets, liquid cooling and hardware vendor support processes
• Fine-tuning, LoRA, quantization and model evaluation methodology
• Czech language

What we offer

• The option to get paid in bitcoin
• Flexible working hours
• A budget for professional development, including training programs, courses and workshops of your choice
• Renovated offices with a gym, massages, football table, billiards, PlayStation and 3D printer, plus free on-site parking
• A MultiSport card, company mobile phone tariff, yoga and fitness classes

About Trezor

Trezor is a leading company in crypto security and the inventor of the world's first hardware wallet. The company is open-source and emphasizes transparency, trust and security in its operations.