Work with the lab
We take on a few applied projects a year, with teams whose problems overlap with our research.
- Evaluations
- Building the benchmark that tells you whether your system works.
- Agents
- Architecture, harnesses and supervision for agents that run unattended.
- Model selection and deployment
- Choosing the smallest model that does the job, and running it well.
- Inference cost and latency
- Finding the decisions that do not need a frontier model.
- Human and AI interaction
- Interfaces where people and models share work without friction.
- Open-source strategy
- Releasing tools and models in a way that people adopt.
Send a few lines about what you are building and where it is stuck. If it is a fit we will reply with questions, not a deck.