EU remote
AI/ML Engineer
About this role
[https://app.ashbyhq.com/api/images/user-content/1246dead-d30e-4c70-a1d3-ef7e5023d542/bad250f8-206a-41f4-9698-287303b69ce3/keyart%20supercell.png] OUR AI TEAM IS LOOKING FOR AN INDIVIDUAL WHO IS EXCEPTIONAL AT THE APPLIED END OF THE CRAFT: TAKING MODERN MODELS AND TURNING THEM INTO FEATURES AND TOOLS THAT HOLD UP IN PRODUCTION, AT SUPERCELL SCALE, UNDER REAL DEADLINES. The challenge at the centre of this role is the distance between a demo and a product.
A model that works in a notebook is not a feature that works for hundreds of millions of players, or a tool a game team trusts with their daily work. Help us Close that gap: the evaluation that tells you it's actually good, the latency and cost envelope that makes it viable, the failure modes you find before players do, and the plumbing that makes it boring to operate. We're not looking for someone who wants to train a frontier model.
We're looking for someone who gets unreasonable results out of the ones that exist. We are looking for someone to build and ship: AI features in live games, and the internal tools that give superpowers to Supercellians. Some of that work will take you inside a game or central team for a while, close enough to their craft to build the right thing with them. In our team, you'll work closely with our AI Platform Engineers, AI/ML engineers, Clients Programmers, Cloud Engineers.
You’ll also collaborate with other teams including Data, QA, Engine and game teams. [https://app.ashbyhq.com/api/images/user-content/1246dead-d30e-4c70-a1d3-ef7e5023d542/7a053edc-bf2e-4db4-85ee-5630e2b5d305/banner_narrow_customs.jpg] WHAT YOU'LL BE DOING - Build and ship AI features and systems end to end into live games and internal tools, at production quality and under real deadlines. - Do the applied engineering that makes AI work in practice.
- Own evaluation as a first-class deliverable. Write and maintain the benchmarks, evals, and regression suites that tell you whether something is genuinely good, across the modalities you're working in. - Fine-tune, adapt, and distill models where it pays off, and have the judgment to know when a better prompt, better retrieval, or a smaller model beats training anything. - Make sound trade-offs on inference, including on-device where it matters, against real parameter, latency, and battery targets, without losing the feel players expect.