Applied Scientist Intern - Trust and Safety (Multimodal Foundation Model) - Global Frontier Tech Recruitment Program - 2027 Start (PhD)

What you'll do

• Research and develop MoE-based multimodal safety models for cross-modal understanding, reasoning, and generation.
• Train and improve moderation agents’ multi-step decision-making, tool use, and context assembly using reinforcement learning.
• Address model generalization and adversarial robustness across languages and moderation scenarios.

What they ask for

• Currently pursuing a PhD in Computer Science, Data Science, Artificial Intelligence, or a related field.
• Strong understanding of cutting-edge LLM research.
• Proficiency in Python, Rust, or C++, and experience with deep learning frameworks.
• Strong understanding of distributed computing and performance tuning for training, fine-tuning, or inference.

Nice to have

• Research publications or contributions to the LLM community.
• Experience with inference tuning and acceleration, AI accelerators, large-scale AI networks, AI system evaluation, or LLM agent development.

Tools

Python, Rust, C++, pytorch, deepspeed, megatron, vllm, GRPO/PPO

AI focus

Research and development of multimodal foundation models and reinforcement-learning-driven agentic moderation systems, including MoE training, cross-modal reasoning, synthetic data generation, tool collaboration, and adversarial robustness.