VP AI Engineering — Pre-Training

About the Role

You’ve likely scaled systems most people only read about.

We’re entering a phase where infrastructure decisions directly shape model capability — not just cost or reliability. The next gains won’t come from brute force alone, but from better systems, sharper tradeoffs, and sound judgment under real constraints.

As VP AI Engineering (Pre-Training), you’ll define how large-scale training actually works here — from compute strategy and data movement to iteration speed. This role is for someone who has been close to the metal, seen what breaks at scale, and knows how to design systems that hold up under pressure.

Infrastructure in this role isn’t a support function. It’s a first-order driver of model outcomes. You’ll be accountable for whether training runs succeed, stall, or fail — and for the decisions behind those outcomes.

This is a role for builders who have owned training systems end-to-end, not for abstract platform or governance leadership.

What You Will Lead

You will own the entire pre-training foundation, including:

• Large-scale compute strategy (GPU clusters in the thousands)

• Training orchestration, throughput, and efficiency

• PB-scale data pipelines, storage, and data movement

• Reliability and performance of long-running training workloads

• Infrastructure decisions that directly impact model velocity and quality

This is not a “keep the lights on” role. You’ll be expected to:

• Make hard architectural calls with incomplete information

• Decide where to invest, where to simplify, and where to say no

• Work closely with modeling and post-training leaders to remove system-level constraints

• Stay close enough to the systems to know when theory diverges from reality

Many of the decisions you’ll make here are difficult to reverse. You’ll commit to architectures, tooling, and operating models that must hold up in production training runs measured in weeks — not just in design documents.

You’ll build a lean team around you, but this role succeeds through clarity and leverage, not headcount.

What You Bring

You’ve likely done several of the following already:

• Designed, operated, or materially evolved large GPU training clusters where you were accountable for throughput, failure modes, and iteration speed — not just budget or vendor relationships

• Built or scaled training infrastructure used by advanced ML teams in production

• Worked with data pipelines measured in petabytes, not terabytes

• Balanced speed, cost, and reliability under real delivery pressure

• Made architectural decisions that held up months or years later

You’ve been close enough to the system to know where theoretical efficiency breaks down in practice — and you’ve adjusted accordingly.

We care less about where you worked and more about what you personally owned, especially when things didn’t go to plan.

Why This Is Different

• Fewer layers between decision and execution

• Direct influence on model capability, not just infrastructure metrics

• A system still early enough to bend — but serious enough to matter

• A founder-led environment where clarity, speed, and judgment outweigh process

• Meaningful equity and real ownership — not cosmetic leadership

If you’ve ever wanted to apply everything you’ve learned without the inertia of a massive organization, this is that moment.

What You Can Expect from Us

• Become part of an AI revolution!

• 30 days of paid vacation

• Access to a variety of fitness & wellness offerings via Wellhub

• Mental health support through nilo.health

• JobRad® Bike Lease

• Substantially subsidized company pension plan for your future security

• Subsidized Germany-wide transportation ticket

• Budget for additional technical equipment

• Flexible working hours for better work-life balance and hybrid working model

• Virtual Stock Option Plan