Working at the frontier of AI
Our researchers, scientists, and engineers push the frontier of training, alignment, agents, and deployment — and bring it to production with the teams we partner with.
Where we focus our work.
Pre-training & scaling
Training frontier models efficiently — data curation, architectures, and scaling behavior on real hardware.
Post-training & alignment
Supervised fine-tuning, RLHF, and preference optimization to align models with human and policy goals.
Agents & tool use
Teaching models to plan, use tools, and complete multi-step tasks reliably and safely.
Evaluation & interpretability
Building rigorous evaluations and understanding model behavior so progress is measurable and trustworthy.
Efficient inference & serving
Pushing throughput and latency to hardware limits for dependable production deployment.
Frontier research, held to a production bar.
Evaluation-driven
Every result is grounded in rigorous, task-specific evaluation — not vibes.
Production-grounded
We research against real workloads, so what works in the lab works in the field.
Open collaboration
We work alongside your researchers and engineers, sharing methods and tooling.
Collaborate with our researchers — or join them.
Partner with us on a research engagement, or help build the frontier as part of the team.