Founding ML Researcher
- Team
- Research
- Openings
- 1
- Location
- Munich
- Workplace
- In person
- Type
- Full time
The role
Build the models that learn how a company works, from the first experiment to production.
Ambient records how experts work across every application on their desktops. Multiplayer joins that work across a company into one continually learned model. The research here decides how that model learns from each new day of work and how every outcome traces back to the steps that produced it.
Each result ships into deployments at companies in regulated, IP-heavy industries, with confidential inference in secure enclaves and the data kept inside the company.
The work
- Train vision language models that understand desktop work in any application, from CAD and Excel to internal tools.
- Build continual learning and parametric memory, so a company’s model keeps learning from new work without forgetting.
- Link delayed outcomes to the work that earned them and train value functions on that signal.
- Turn a company’s real work into RL environments, verifiers and evals that measure any model on it.
- Run experiments fast on our own GPU clusters and carry the ones that work into production.
The bar
- Research shipped as working systems: papers, open-source code or models in production.
- Deep command of post-training, reinforcement learning and multimodal models.
- Agent-native: swarms of coding agents are the daily tools.
- Ownership and pace in a small team, from idea to deployed result.
- Nice to have: continual learning, memory in weights, long-horizon credit assignment or computer-use agents.