Senior AI Infrastructure Engineer

About the Role

This is the last infrastructure role the CTO still owns himself, and the #1 hiring priority for the company. You'll own the core of Mage's backends — the generation services for image, video, and audio — and the integrations that bring new models to users. AI velocity is relentless: new models and provider APIs drop daily, and the team needs someone who can get the latest and greatest into production within 24–48 hours, because a day is a lifetime in AI. High ownership, fast shipping, real users.

Key Responsibilities

• Own and operate the core backend generation services (image, video, audio).

• Integrate new models continuously — open-source (Stable Diffusion, SDXL, diffusers) and closed/third-party provider APIs (e.g. Seedance 2, Krea 2) — and ship them to users within 24–48h of release.

• Design clean, consistent APIs over a fast-changing set of models.

• Deploy and operate GPU inference: cold starts, concurrency, autoscaling, latency, and cost.

• Build production orchestration across inference, post-processing, and media pipelines, with retries, moderation, and error handling.

• Run it all on Google Cloud.

Requirements

• A

  Excellent Python and strong API design (e.g. FastAPI) for production backends.

• B

  Deep command of backend fundamentals — how the internet and scalable backends are actually structured.

• C

  A live, shipped project that proves ability ("show, don't tell") — owned a feature from ideation through to production serving real users (hundreds to millions); solo or on a team is fine, as long as they can speak to it.

• D

  Cloud and containerized deployment experience — Google Cloud (Cloud Run, GCS) and Docker.

Bonus Skills

• GPU deployment and inference at production scale (cold starts, concurrency, autoscaling, latency, cost) — Modal or equivalent.

• Open-source model hacking — hands-on with or contributions to vLLM, diffusers, or Hugging Face projects (a major green flag given Mage's open-source genesis).

• Familiarity with the generative-AI ecosystem: image/video/audio models and inference providers.

• PyTorch, Redis.

Ideal Background

• Strongest signal — consumer + startup: engineers who've shipped fast at consumer products and at early-stage startups. Experience across both consumer and startup is the ideal profile.

• Open-source / model-hacking background (vLLM, diffusers, Hugging Face) is a major green flag.

• Anywhere that ships fast — industry-agnostic; what matters is velocity and ownership.

• Calibration: the ideal hit is a young-at-heart, entrepreneurial, ships-fast generalist who treats AI as central to how they build and ramps on new tools rapidly (the profile of a current standout teammate).