Full-time
Researcher, RL
London, England, United Kingdom · Hybrid
£120,000 - £240,000
Researcher, Reinforcement Learning for Agents
Agents that do well on benchmarks often struggle once they meet real tasks, and most of them don’t get any better with use. This team is working on that gap. They want agents that learn from real work: their failures become training signal, and each new version is measurably better than the last.
They’re an early-stage, well-funded AI research startup. The founders are researchers and operators from leading AI labs, and they’re already working with enterprise customers. The research team is small, so you’d help set the direction rather than inherit it.
The work runs across the whole training stack. You don’t need experience in all of it:
Continual learning, so improvements build up over time without breaking what already works
Post-training with RL and preference optimisation (PPO, GRPO, DPO and beyond)
Environments, simulators and synthetic task generation for long-horizon, multi-step work
Curricula that adapt as agents improve
Verifiers, benchmarks and evals that show whether training gains hold up in real use
Fast, efficient RL training loops: rollouts, rewards, policy updates, GPU utilisation and training-inference bottlenecks, using tools such as PyTorch, vLLM, SGLang and Ray
Some people will lean towards environments and evals, others towards making RL train efficiently at scale, and some will do both. Strong candidates have usually post-trained language models or built RL systems that had to work outside a paper, and they’re comfortable making research decisions with incomplete information.
📍 London, hybrid
💰 Up to circa £240K base plus equity at Staff level; other levels depend on experience
✅ Well funded by top-tier VCs | Founders from leading AI labs
If it sounds relevant, we’d be happy to talk it through.
All applicants will receive a response.
Salary
£120,000 - £240,000
Location
London, England, United Kingdom · Hybrid
Contract
Full-time
Your consultant
Marc Powell

Full-time
Researcher, RL
London, England, United Kingdom · Hybrid
£120,000 - £240,000
Researcher, Reinforcement Learning for Agents
Agents that do well on benchmarks often struggle once they meet real tasks, and most of them don’t get any better with use. This team is working on that gap. They want agents that learn from real work: their failures become training signal, and each new version is measurably better than the last.
They’re an early-stage, well-funded AI research startup. The founders are researchers and operators from leading AI labs, and they’re already working with enterprise customers. The research team is small, so you’d help set the direction rather than inherit it.
The work runs across the whole training stack. You don’t need experience in all of it:
Continual learning, so improvements build up over time without breaking what already works
Post-training with RL and preference optimisation (PPO, GRPO, DPO and beyond)
Environments, simulators and synthetic task generation for long-horizon, multi-step work
Curricula that adapt as agents improve
Verifiers, benchmarks and evals that show whether training gains hold up in real use
Fast, efficient RL training loops: rollouts, rewards, policy updates, GPU utilisation and training-inference bottlenecks, using tools such as PyTorch, vLLM, SGLang and Ray
Some people will lean towards environments and evals, others towards making RL train efficiently at scale, and some will do both. Strong candidates have usually post-trained language models or built RL systems that had to work outside a paper, and they’re comfortable making research decisions with incomplete information.
📍 London, hybrid
💰 Up to circa £240K base plus equity at Staff level; other levels depend on experience
✅ Well funded by top-tier VCs | Founders from leading AI labs
If it sounds relevant, we’d be happy to talk it through.
All applicants will receive a response.
Your consultant
Marc Powell

Full-time
Researcher, RL
London, England, United Kingdom · Hybrid
£120,000 - £240,000
Researcher, Reinforcement Learning for Agents
Agents that do well on benchmarks often struggle once they meet real tasks, and most of them don’t get any better with use. This team is working on that gap. They want agents that learn from real work: their failures become training signal, and each new version is measurably better than the last.
They’re an early-stage, well-funded AI research startup. The founders are researchers and operators from leading AI labs, and they’re already working with enterprise customers. The research team is small, so you’d help set the direction rather than inherit it.
The work runs across the whole training stack. You don’t need experience in all of it:
Continual learning, so improvements build up over time without breaking what already works
Post-training with RL and preference optimisation (PPO, GRPO, DPO and beyond)
Environments, simulators and synthetic task generation for long-horizon, multi-step work
Curricula that adapt as agents improve
Verifiers, benchmarks and evals that show whether training gains hold up in real use
Fast, efficient RL training loops: rollouts, rewards, policy updates, GPU utilisation and training-inference bottlenecks, using tools such as PyTorch, vLLM, SGLang and Ray
Some people will lean towards environments and evals, others towards making RL train efficiently at scale, and some will do both. Strong candidates have usually post-trained language models or built RL systems that had to work outside a paper, and they’re comfortable making research decisions with incomplete information.
📍 London, hybrid
💰 Up to circa £240K base plus equity at Staff level; other levels depend on experience
✅ Well funded by top-tier VCs | Founders from leading AI labs
If it sounds relevant, we’d be happy to talk it through.
All applicants will receive a response.
Your consultant
Marc Powell


