Senior Research Scientist - Multimodal Agents

You will lead the research and engineering of multimodal agent systems, focusing on planning, tool use, memory, and retrieval to solve real design, vision, and language tasks.
Scale post-training and reinforcement learning across distributed PyTorch pipelines, ensuring efficient, reproducible training of large models and MoE architectures.
Design and implement reward models and learning loops (RLHF/RLAIF, preference modeling, DPO/IPO-style objectives, offline/online RL, curriculum learning) for multi-step reasoning.
Create simulation environments and evaluation suites to surface failure modes (planning errors, tool use brittleness, hallucination, unsafe actions) and turn them into measurable targets.
Lead rigorous evaluation: offline suites, online experiments, and data-backed conclusions; collaborate with product, safety, and platform teams to ship reliable features.
Mentor teammates, present results, and contribute to the research community while upholding high standards of reproducibility and safety.