Job description

ABOUT VAST.AI Vast.ai runs one of the largest GPU marketplaces in the world. Over 20,000 GPUs, from RTX 4090s up to B300s, rented by more than 25,000 customers a month who train, fine-tune, and serve AI models on them. We are profitable, the team is flat, and we ship quickly. Your application goes straight to the hiring team. THE ROLE This is an engineering job that happens in public. You will be our heaviest user. Most of your week goes to renting GPUs on Vast and building things on them. Serving open source models with vLLM and SGLang. Fine-tuning. Running ComfyUI pipelines. Keeping live inference endpoints up, including token endpoints on markets like OpenRouter. Writing the example repos, Docker templates, and benchmarks that other developers copy. The rest of your week is spent showing your work. Short videos, technical write-ups, answers in Discord and on Reddit, hackathons. You are not building the Vast product. You are using it harder than any of our customers, in the open, and reporting back everything that breaks. If you want a content calendar and a campaign plan, this is the wrong job. If you already spin up GPUs on weekends because you want to see whether the new model actually runs, keep reading. WHAT YOU'LL DO - Run real workloads on Vast every week. Deploy, fine-tune, and serve open source models, and keep live endpoints running. - Build public example repos, Docker templates, and benchmarks that developers can clone and use. - Turn what you build into teaching: guides, short videos, talks, livestreams. - Be the most technically credible voice in our Discord, on Reddit, and on X. - Represent Vast at hackathons and conferences, roughly one trip a month. - Report the friction, the breakage, and the missing docs straight to engineering and product. WHAT WE NEED TO SEE - You can take an open source model from Hugging Face to a running endpoint on rented GPUs with nobody helping you. Linux, Docker, Python, SSH. - You can debug the GPU stack. Driver and CUDA mismatches, out of memory errors, and multi-GPU configuration including NCCL and topology issues do not scare you. - Public proof of work you have shipped. A GitHub profile, technical writing, or live projects. We will ask you to walk us through code you wrote and explain the parts that went wrong. - You work without a roadmap and prefer it that way. Nobody here will hand you a task list. - You can explain a hard thing clearly, in writing and out loud. NICE TO HAVE - Content or side projects that found a real audience. - Previous developer relations, community, or customer-facing work. Useful here, but it is a bonus and not the job. - vLLM or SGLang internals, distributed training with FSDP or DeepSpeed, CUDA. - Time at a cloud, GPU, AI infrastructure, or developer tools company. FIRST 90 DAYS - Ship several public example projects on Vast, including at least one live serving endpoint. - Publish benchmarks or guides that developers actually use and pass around. - Become the recognizable technical voice in our community channels. - Represent Vast at a sponsored hackathon. COMPENSATION & BENEFITS $160,000 to $200,000 base, plus equity and bonus. Health, dental, vision, and life insurance. 401(k) with company match. Meals on site. Travel and conference budget. This role is on site in San Francisco or Los Angeles.

Requirements

Must-haves

This is an engineering job that happens in public. You will be our heaviest user.
If you want a content calendar and a campaign plan, this is the wrong job. If you already spin up GPUs on weekends because you want to see whether the new model actually runs, keep reading.
Run real workloads on Vast every week. Deploy, fine-tune, and serve open source models, and keep live endpoints running.
Build public example repos, Docker templates, and benchmarks that developers can clone and use.
Turn what you build into teaching: guides, short videos, talks, livestreams.
Be the most technically credible voice in our Discord, on Reddit, and on X.
Represent Vast at hackathons and conferences, roughly one trip a month.
Report the friction, the breakage, and the missing docs straight to engineering and product.
You can take an open source model from Hugging Face to a running endpoint on rented GPUs with nobody helping you. Linux, Docker, Python, SSH.
You can debug the GPU stack. Driver and CUDA mismatches, out of memory errors, and multi-GPU configuration including NCCL and topology issues do not scare you.
Public proof of work you have shipped. A GitHub profile, technical writing, or live projects. We will ask you to walk us through code you wrote and explain the parts that went wrong.
You work without a roadmap and prefer it that way. Nobody here will hand you a task list.
You can explain a hard thing clearly, in writing and out loud.
Ship several public example projects on Vast, including at least one live serving endpoint.
Publish benchmarks or guides that developers actually use and pass around.
Become the recognizable technical voice in our community channels.
Represent Vast at a sponsored hackathon.

Nice-to-haves

Content or side projects that found a real audience.
Previous developer relations, community, or customer-facing work. Useful here, but it is a bonus and not the job.
vLLM or SGLang internals, distributed training with FSDP or DeepSpeed, CUDA.
Time at a cloud, GPU, AI infrastructure, or developer tools company.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

aidockerrestpython

Platform: ashby
Effort: quick

Benefits

Benefits detected in the listing.

Dental

Vision

401(K)