ML Researcher, Foundation Models

What you'll do

• Pre-train and develop mid-training approaches for foundation models to bridge general capability and task-specific performance.
• Implement reinforcement learning techniques like RLHF, GRPO, and PPO to improve reasoning and coding capabilities.
• Develop autonomous coding agents capable of operating in real repositories and handling multi-step workflows.
• Design reward modeling systems to resist reward hacking and handle sparse or delayed rewards in agentic settings.
• Research scaling laws for reinforcement learning compute allocation and progressive alignment across capability stages.
• Train models to manage their own context during long-horizon tasks using techniques like sparse attention or compression.
• Optimize foundation models specifically for Apple silicon and integrated OS experiences.
• Apply distillation techniques to combine independently optimized capabilities into a single, high-performing model.

What we're looking for

• PhD or equivalent practical experience in Computer Science or a related technical field.
• Demonstrated expertise in deep learning with publications at top ML/NLP conferences or a track record of applying techniques to products.
• Proficiency in Python and at least one deep learning toolkit such as JAX, PyTorch, or TensorFlow.
• Experience in reinforcement learning for LLMs including RLHF, GRPO, PPO, reward modeling, and scaling laws.
• Expertise in code generation, repository-level code understanding, and agentic coding systems.
• Experience with agentic systems involving multi-turn RL, tool-use planning, and long-horizon task execution.
• Knowledge of distillation and alignment techniques such as on-policy or reward-tilted distillation.
• Familiarity with long context and efficiency techniques like sparse attention and context compression.