Research
We believe the best research comes from a cycle of curiosity, building, feedback and iteration. Everyone in the lab follows this loop.
Long-context architectures, retrieval, and persistent memory systems.
Agent architectures, tool use, planning, and multi-agent coordination for autonomous systems.
Alignment, RLHF, instruction tuning and techniques that improve model behavior.
Embodied intelligence, control, manipulation and real-world robotics.
Building fast, reliable and cost-efficient inference systems at scale.
Rigorous evaluation frameworks and benchmarks to measure model and agent behavior reliably.
Compressing large models into smaller, faster ones without losing capability.
Ranking, retrieval, and personalization systems that connect users to what matters.
Text-to-speech, speech-to-text, and the models and pipelines built around them.
Accelerators, FPGAs, and the hardware systems that make AI fast and efficient.