About the role
You work on the intelligence itself, on a team where research that does not reach a user is unfinished. The questions are shaped by our constraint: the model often runs on hardware the person owns, under a consent boundary, which rules out a lot of easy answers and makes the remaining ones interesting.
What you’ll do
- Push the frontier on personal AI: knowledge/world models, memory, retrieval, evals
- Build the data and systems substrate - efficient, sovereign, on-device and edge-first
- Turn research into shipped capability with real latency, accuracy, and relevance budgets
- Keep the human in control: consent-native design, interpretability, and honest evaluation
- Publish and teach in the open where we can; collaborate with our university labs
What we need to see
- Real research contribution: published work, or a shipped system whose core idea was yours
- You implement your own ideas to a production standard rather than handing them over
- Rigor about evaluation: you can tell a genuine improvement from a better-looking benchmark
- Depth in at least one of reasoning, memory, retrieval, efficiency, or agent architecture
Nice to have
- Quantisation, distillation, or on-device inference, since that is where our constraint bites
- Open-weights work, or contributions to an open model ecosystem
- Privacy-preserving machine learning
What winning looks like
- Eval and benchmark gains that reach the product
- Research shipped with real latency and accuracy budgets
- Reproducibility and honest evaluation
Where and how we work
In the office together five days a week, in any of these cities. Remote-friendly around your family, arranged one person at a time.