AI Research Engineer (Agentic Reinforcement Learning)

About the role

This position is listed on behalf of a partner company, who manages all applications and next steps. Our partner is looking for an AI Research Engineer (Agentic Reinforcement Learning) based in United States.

As an AI Research Engineer specializing in Agentic Reinforcement Learning, you will drive research and innovation in next-generation intelligent systems. You will develop advanced reinforcement learning techniques that improve decision-making, adaptive behavior, and performance across multimodal AI architectures. Working with text, images, audio, and other data modalities, you will help build models capable of solving complex, real-world challenges. Your work will involve designing scalable training infrastructure, optimizing reward mechanisms, and improving the reliability of autonomous AI agents. You will explore cutting-edge approaches involving diffusion-based generative models, autoregressive architectures, and unified multimodal frameworks. Through rigorous experimentation and benchmarking, you will push the boundaries of model performance, efficiency, and generalization. This fully remote opportunity offers the chance to contribute to impactful AI research within a fast-moving, globally distributed technology environment.

Accountabilities In this role, you will conduct advanced research and develop scalable reinforcement learning systems for multimodal AI applications. Your main responsibilities will include:

-

Research and develop reinforcement learning algorithms for multimodal models, including diffusion-based approaches for image generation, autoregressive models for multimodal understanding, and unified architectures integrating multiple modalities.

-

Design and implement scalable reinforcement learning infrastructure supporting distributed training across complex multimodal systems.

-

Develop and optimize reward modeling strategies to improve training stability, align model behavior with desired outcomes, and prevent reward hacking and related failure modes.

-

Create, curate, and maintain multimodal simulation environments and training datasets to support robust model development, evaluation, and benchmarking.

-

Design and implement rigorous evaluation protocols to measure model performance, monitor progress against established baselines, and validate research improvements.

-

Analyze and optimize policy performance across different modalities by identifying bottlenecks in training, credit assignment, reward optimization, and cross-modal alignment.

-

Investigate emerging reinforcement learning paradigms that enable AI agents to learn more effectively from environmental feedback.

-

Improve model exploration, decision-making, and adaptive behavior to achieve state-of-the-art performance in dynamic, real-world environments.

-

Conduct large-scale experiments and translate empirical findings into practical algorithmic improvements.