Applied Scientist Intern - Trust and Safety (Multimodal Foundation Model) - Global Frontier Tech Recruitment Program - 2027 Start (PhD)
Core
Research and build multimodal foundation models and agentic moderation systems to protect users from negative content using state-of-the-art machine learning.
Role type
PhD-level Applied Scientist (Research & Engineering)
Builds
Multimodal safety foundation models and RL-driven agentic moderation systems
Domain
AI Safety, Content Moderation, Multimodal AI, Reinforcement Learning
Deliverable
production ML models
Required skills
Multimodal foundation model training, MoE architecture optimization, Reinforcement Learning (GRPO/PPO), Agentic decision-making, Cross-modal alignment, Distributed computing, Python, Rust, C++, Deep learning frameworks (PyTorch)
Preferred skills
Published research papers, Inference tuning and acceleration, GPU/AI accelerator expertise, LLM application development, GraphRAG strategies
Technologies
PyTorch, DeepSpeed, Megatron, vLLM, LangChain, MCP, GRPO, PPO
Responsibilities
Train large-scale sparse MoE models for multimodal understanding and generation, Develop RL-based agents for multi-step reasoning and tool collaboration, Engineer dynamic context assembly and heterogeneous evidence fusion, Design reward signals for adversarial robustness and few-shot scenarios
Seniority
PhD Candidate (Research Intern)
