Senior AI Researcher
Core
Lead post-training evaluation, red-teaming, and reinforcement learning (RL) gym audits on open-weight models to establish benchmarking methodologies and measure security capabilities against threats like Indirect Prompt Injections.
Role type
Senior AI Researcher (RL & Security)
Builds
Post-training evaluation pipelines, RL environments, and security benchmarking frameworks for open-weight LLMs.
Domain
AI Safety, Machine Learning, Cybersecurity
Deliverable
production ML models
Required skills
Reinforcement Learning (GRPO, PPO), LLM Red-teaming, Indirect Prompt Injection (IPI) defense, PyTorch, Docker, Distributed Training, Reward Diagnostics, Trace Analysis
Preferred skills
Harbor format experience, Agentic workflow evaluation, Open-weight model evaluation (Llama, Mistral)
Technologies
PyTorch, Docker, GRPO, PPO, Harbor
Responsibilities
Execute post-training runs using RL algorithms on open-weight models; Analyze live loss curves and rollout traces to identify reward hacking; Review tasks and multi-turn environments for realism and threat model accuracy; Generate comprehensive evaluation cards detailing performance uplift and failure modes; Integrate dockerized environments into internal training frameworks.
Seniority
Senior, hands-on IC
