Agentic RL Head

UMI TECH; Singapore; On-site; SGD 150,000 - 210,000; Full time

Job summary

UMI TECH seeks a Senior or Head Agentic Reinforcement Learning lead to drive next-generation AI agents capable of autonomously executing long-horizon tasks across virtual environments. You will steer RL post-training research, tackling GRPO/PPO, trajectory-based training, and sandbox integration to push agent autonomy and reliability.

You will lead a small team, deliver end-to-end LLM post-training projects, and work closely with evaluation to maximize task success rates in complex environments.

Qualifications

• Master's degree or above with hands-on RL experience.
• Experience leading a small team on end-to-end LLM post-training projects.
• Familiarity with agent-based tasks and long-horizon task objectives.

Responsibilities

• Conduct RL training experiments for long-horizon tasks and tune hyperparameters.
• Manage agent trajectory data and sandbox environment integration for training pipelines.
• Collaborate with eval teams to validate model capabilities and improve success rates on long-horizon tasks.

Skills

Reinforcement Learning

RLHF

RLVR

Agent-based projects

LLM training

Education

Master's degree or above

Tools

GRPO

PPO

veRL

OpenRLHF

Job description

We are looking for a Senior or Head Agentic Reinforcement Learning (RL) to lead the development of next-generation AI agents capable of autonomously executing complex, long-horizon tasks across virtual environments.

This is an exciting opportunity to drive cutting-edge LLM post-training and Agent RL, working on advanced techniques such as GRPO/PPO, trajectory-based training, and sandbox environment integration. You will lead a team to push the boundaries of agent autonomy, reliability, and task completion, turning advanced AI research into real-world capabilities.

Job Responsibilities

Conduct reinforcement learning (RL) training experiments for long-horizon tasks, including hyperparameter tuning, performance analysis, and iterative optimization.

Manage agent trajectory data feedback and integration into the training pipeline, as well as coordinate and optimize sandbox environment integration.

Work closely with the evaluation team to validate model capabilities, analyze evaluation results, and continuously improve the model's success rate on long-horizon tasks.

Requirements

Master's degree or above, with at least 3 years of hands-on experience in Reinforcement Learning (RL), Reinforcement Learning from Human Feedback (RLHF), or Reinforcement Learning with Verifiable Rewards (RLVR). Experience in Agent-related projects is preferred.

Familiarity with mainstream RL algorithms such as GRPO and PPO, as well as LLM training frameworks such as veRL and OpenRLHF.

Experience leading a small team, with a proven track record of delivering end-to-end LLM post-training projects.