Astrocade is an AI technology company building super-intelligence for the interactive world, delivered through a breakout consumer platform.

Today's frontier models can write, reason, and generate video. None can build a world you can step into and act on. Interactive world models are one of the defining problems on the path to ASI, and the foundation of the next great consumer platform. Interactivity is the holy grail of content: not something you watch, but something that responds to you. We're further ahead than anyone: the only team building the intelligence and the platform together, and both are already live.

Our social UGC platform lets anyone create and play interactive experiences through natural language. And it's one of the fastest-growing consumer products in the world, with over 30 million users in just months since launch. Creators bring to life games, companions that remember and grow with them, learning experiences that adapt like a great tutor, and social worlds built with friends. The platform also powers our research with data no video dataset contains: millions of creators prompting and refining gives us paired data of human intent and the worlds that fulfill it, and every play session produces ground-truth, action-labeled interaction data at scale. All of it flows back into our models, and research ships to real players in days, spinning a flywheel no one else can build.

We are a small team from Google Research, Meta, FAIR, Character.ai, and Stanford, led alongside Fei-Fei Li, with frontier-scale compute and direct ownership of core model problems. We are backed by Sequoia Capital, Google, NVIDIA, and luminaries including Eric Schmidt and Jerry Yang.

The next foundational AI company will be built where world models meet a billion players. We're the rare place where frontier research and a breakout consumer platform are one company, and where your work ships to millions, not to a paper. Come build it with us.

About the Role

Every experience on Astrocade starts as an idea from a creator. An agentic AI system turns that idea into something playable. How good that experience turns out to be is our north-star metric. Your job is to move that metric.

As an ML Engineer on our AI team, you'll own how our creation system behaves, where it fails, and how it gets better. You'll dig into real creator sessions to find where generation breaks down, build the evals that tell us whether a change actually helped, and design the agentic systems behind the next version of creation on the platform. This is applied ML with a direct line to product. What you ship shows up in the games millions of players see.

We're a small team of about 20 engineers, and nobody here stays in a lane. You'll work across models, agents, data pipelines, and infrastructure as the problem requires, and you'll collaborate directly with our CTO and co-founders.

What You'll Do

Own the quality of AI-generated content end to end, including model behavior, agent orchestration, prompting, tool use, and the failure modes in between

Build evaluation and insight systems on production creator data that measure failure rates, quality, and where the system is strong or weak, and turn those findings into concrete improvements

Design and build agentic ML systems that take a creator's idea and reliably produce a great, playable game

Debug hard problems across the stack, from a bad generation down to the data, the model, or the pipeline that caused it

Run experiments quickly, measure rigorously, and ship what works

Help define how the AI team works as it grows, including tooling, eval standards, and technical direction

You'd Be a Great Fit If You

Have 3–7+ years of experience in ML engineering or applied ML, with a track record of shipping models or AI systems to real users

Have built LLM-powered or agentic systems in productionTreat evals as a core part of the job: you've designed metrics and eval sets and know when a number is lying to you

Have strong CS fundamentals and write clean, efficient code (Python-heavy)