Member of Technical Staff Research Scientist
Push the frontier of reinforcement learning for long-horizon work
London, UK Full-time Research
Context
Reinforcement learning is how language models learn to reason, use tools, and work autonomously over long horizons. But RL for long-horizon agents presents new challenges, including off-policyness of long trajectories, sparse credit assignment, and training instability.
At General Reasoning, RL research is core to what we do. We train agents on environments hosted on OpenReward and need researchers who can design and implement novel RL methods, run experiments at scale, and push the capabilities of long-horizon agents. Before GR, our team previously worked on some of the leading open-source language model efforts and early reinforcement learning work on language models.
About this Role
As a Research Scientist, you'll design, implement, and evaluate agentic reinforcement learning methods for language models. You'll work across the full research loop - formulating problems, building training pipelines, running experiments, analysing results, and publishing your work. Your research will directly shape how agents are trained on the OpenReward platform.
Design and implement novel RL algorithms and training methodologies for long-horizon language model agents
Develop and iterate on reward signals, shaping methods, and credit assignment strategies for long-horizon tasks
Run RL training experiments at scale, analyse agent behaviour, and debug failure modes such as reward hacking and training instability
Build and maintain research infrastructure: training pipelines, experiment tracking, and evaluation harnesses
Publish papers and engage with the broader RL and LLM research community
Collaborate closely with the engineering team to bring research innovations into production training
Requirements
We're looking for exceptional individuals who can operate at the frontier of AI research and engineering.
Obsessive attention to detail, high intensity and focus
Evidence of extraordinary accomplishment (either in work or early life)
Kind, self-aware and collaborative team player
Strong background in reinforcement learning: policy gradient methods, value-based methods, PPO, or related algorithms
Experience training RL agents and iterating on reward design, with strong intuitions for what works and why
Proficiency in Python and ML frameworks (PyTorch or JAX) with the ability to implement and run experiments independently
Track record of research output: publications, open-source contributions, or impactful technical projects
Willingness to work in-person in London, UK
Nice to Have
Experience applying RL to language models with verifiable rewards (RLVR)
Experience with long-horizon or multi-step RL tasks and the specific challenges they present
Experience with distributed training infrastructure and running experiments at scale on GPU clusters
Background in LLM training methodologies (pre-training, post-training, fine-tuning)
Experience working at a frontier lab on critical path research
About General Reasoning
General Reasoning (GR) was founded by ex-frontier lab researchers who were at the forefront of the LLM revolution. We've built models, products and evaluations used by hundreds of millions of people, and have extensive experience across the LLM stack. We are backed by pioneers in the field including the creators of PyTorch and TensorFlow.
We are well-funded and have recently launched OpenReward, the platform for serving RL environments at scale.
#J-18808-Ljbffr