FractalBrain open roles.
Who we are
FractalBrain is building the intelligence layer for a world of continually adapting machines. Our thesis is that human-inspired robotics will drive near-term adoption, while embodied multi-agent systems and adaptive, custom-fit robots will increasingly dominate industry and research needs. The enduring opportunity is intelligence that grows across bodies and coordinated teams, accumulating skills through experience.
What we're building
Today's AI is trained once and then frozen. Everything it knows is fixed before it meets the world, and teaching it something new risks breaking what it already knows.
Brains don't work that way. They keep learning for a lifetime, grow new connections when they need them, and look at the world selectively instead of processing every pixel at full resolution.
How we hire
We hire for how people think. We are a small team of researchers and engineers, some of whom have spent decades on reinforcement learning, planning and multi-agent systems.
We don't care much which frameworks you know. Our core isn't standard deep-learning code, so most of what we use we built ourselves. What we care about is whether you can reason from first principles, tell a real result from a lucky one, and build things that actually work.
The roles below are where we need help most right now.
What you'll work on. Our AI learns continuously at runtime, in dynamic environments, rather than being trained once and frozen. That changes what safety means. You'll build interpretability and monitoring tools for a system that keeps changing, design evaluations that run throughout its lifetime, and study how it recognises when it's out of its depth. You'll work closely with the core team, so what you find shapes the architecture. As the first person in this role, you'll set the lab's safety research agenda and build the safety function as the team grows.
Who you are. You've worked on safety problems such as interpretability, robustness, evals or OOD detection, PhD or equivalent. You've led safety research or a small team, and you're ready to own an area from scratch. Most safety tooling out there assumes a frozen neural net trained with backprop. Ours is neither, so a lot of the time there's no existing method and you'll have to build one. You write clearly and can make a call when the problem isn't well defined.
What you'll work on. You'll build environments and benchmarks for continual multi-agent learning. These are open-ended worlds where goals shift and other agents adapt over time. You'll run experiments in them, find where our agents fail, and work with the core team to fix it. Some of this is research. A lot of it is systems engineering, because our models are large sparse graphs rather than tensors, and making them fast is an open problem.
Who you are. You've worked on reinforcement learning, multi-agent systems or continual learning, and you've shipped code other people relied on. You read papers critically and can tell when a benchmark measures less than it claims. You enjoy open-ended problems and fast iteration, and you don't need an existing framework to get started.
What you'll work on. You'll start with established robotics environments and build the bridge between them and our agents. Sensor streams go into our active-sensing interface, and actions come back out in real time. You'll also build the simulation and test setup that lets us iterate without breaking hardware. Then you'll help answer the question we care about most: what does a robot gain when it can continually learn in the field instead of being trained once? You'll own the robotics roadmap, including which platforms and hardware we bet on, and hire the robotics team over time.
Who you are. You've built robots that worked outside a demo, and you've led robotics projects or teams. You understand control, state estimation and the real-time constraints of physical systems, and you're happy debugging on real hardware. You've used simulators and know where they mislead you and how they can accelerate you. You're comfortable making hardware and vendor decisions. Experience with robot learning helps, but curiosity about a non-standard approach matters more.
What you'll work on. You'll be a thought partner to the CEO and executive team. You'll run company planning, lead strategic projects from problem to delivery, and make sure decisions lead to action. You'll own the pitch deck, budgets, financial projections, and board and investor updates, and support fundraising. As we grow from a research team into a research and product company, you'll oversee the execution of company strategy. Based in London, in the office two days a week to start.
Who you are. You have 5+ years in strategy consulting, product management or a similar role at a venture-backed startup. You build order where there was none, and you've helped brilliant, disorganised leaders become more effective without slowing them down. You write clearly and briefly, can draft in an executive's voice, and are comfortable with financial models, budgets and product metrics. You finish things. Experience in AI, machine learning or robotics helps; deep technical knowledge is a plus, not a requirement.
Excited about our mission and think you have something to contribute, even if these roles aren't a perfect fit? Please reach out anyway as we are always looking for exceptional talent.