Software Engineer (Model Inference)

1 мес. назад

400k–800k USD / yearAustraliaWorldwideSeniorOnsite

cudagpumlmodel inference

Responsible for owning and optimizing the model inference stack, serving AI models with high performance and low latency to millions of users.

At , we're redefining entertainment with AI. We're one of the fastest-growing startups globally, serving 20 million users in our first year . We're an in-person company based in Sydney, Australia.
We're guided by a set of principles that are foundational to our work and drive every decision: User-first: We build things that people want. We invest time to understand our users and focus on adding value instead of extracting value.
High agency, high ownership: We're responsible for the pieces we own, end-to-end. We own every mistake, figure out what went wrong, and fix it. We don't blame anyone or anything else.
Urgency: This is a once-in-a-lifetime opportunity. We prioritize well, find ways to increase leverage, and move at an inspirational pace.

You'll own our model inference stack end-to-end — serving our in-house and open-source models to tens of millions of users at low latency and high throughput, and squeezing every bit of performance out of our GPU fleet.

Competitive compensation with meaningful upside.
A company card for food, coffee, tools, and anything that helps you operate at a high level.
Daily team lunch and dinner at the office.
Unlimited workspace budget. Build your ideal setup.
Real ownership and impact from day one.
Based in Sydney and hiring globally. We sponsor visas and offer relocation assistance to help you make the move.

Ship a high-throughput inference server for our in-house models, serving millions of generations per day at <200ms latency.
Squeeze more out of every GPU with batching, quantization, and custom CUDA kernels .
Test and productionize LoRAs and our in-house models to serve 10s of millions of users.
5+ years of experience building software at scale, with a focus on ML inference or GPU-accelerated systems .
Deep familiarity with GPU inference : batching, quantization, and serving frameworks like vLLM, TensorRT, or Triton.
The ability to get shit done end-to-end . From understanding users → proposing an idea → implementation → iteration.
Hunger to win. This is not going to be easy.
Our overarching philosophy is to raise the ceiling for our best performers, not the floor. We pay based on the value you add to the company.
The range listed for this role is total comp: cash plus equity combined, with superannuation paid on top.
We re-evaluate compensation every 6 months. Bonuses are backward-looking; raises are forward-looking.
A bonus rewards work you've already done, like exceptional output or consistently performing above your level.
A raise reflects that you've moved up. It resets your baseline to match the higher level you're now operating at.