Staff Research Scientist - Reinforcement Learning

📋 Description

• • Design simulation environments and digital twins to model enterprise workflows for AI agent training and evaluation.
• • Post-train large language models using reinforcement learning methods including RLHF, DPO, GRPO, and PPO to align agent behavior with human preferences and task outcomes.
• • Build end-to-end pipelines that convert human-labeled traces and verifiable signals into structured, high-quality training data for reinforcement learning.
• • Architect multi-turn, tool-using LLM agents with closed-loop learning systems that iteratively improve through feedback and environmental interaction.
• • Design reward functions and verifiers that are robust to reward hacking and accurately reflect real-world task success metrics.
• • Set and uphold the technical bar across the research team through architecture reviews, code standards, and engineering best practices.
• • Mentor junior researchers and engineers, driving technical direction through influence rather than authority.
• • Translate cutting-edge research into production-grade systems deployed in enterprise environments such as healthcare, finance, and compliance.
• • Contribute to peer-reviewed publications and present findings at top-tier AI conferences.
• • Collaborate with industry leaders including NVIDIA and Microsoft to advance the state of the art in generative AI and agent systems.
• • Develop and optimize reinforcement learning systems using Gymnasium-based environments with attention to sparse and dense reward structures.
• • Implement and maintain modern RL post-training and rollout-serving libraries such as TRL, veRL, OpenRLHF, and SkyRL in production contexts.
• • Ensure all AI systems are governed, compliant, and trustworthy for enterprise adoption in safety-critical domains.
• • Lead the development of scalable, distributed training pipelines on GPU clusters to support large-scale RL experiments.
• • Apply advanced RL techniques including MDPs, policy gradient methods (PPO, SAC), and temporal difference learning to solve complex enterprise problems.
• • Engineer LLM agents capable of tool use, multi-turn reasoning, and trajectory evaluation with high reliability and interpretability.
• • Maintain strong software engineering discipline: build production pipelines, not just research notebooks, using Python and modern ML tooling.