Research Engineer - Post-Training

Pluralis Research is seeking a Research Engineer for a remote, full-time position focused on post-training for their decentralized Protocol Learning initiative. The role involves developing a reinforcement learning (RL) training loop that integrates rollout ingestion from a geo-distributed inference pipeline, computes rewards, updates policies, and disseminates updated model weights across a network. Candidates should be adept at adapting standard RL algorithms to function effectively in an asynchronous, high-latency environment with partially trusted hardware. This includes addressing challenges such as staleness tolerance and communication-efficient policy updates. The position requires inventiveness in algorithm development and the ability to demonstrate model improvements through evaluations. While the advert does not specify exact qualifications or years of experience, it implies that a strong background in reinforcement learning and decentralized systems is essential. The role is designed for individuals who thrive in innovative, research-driven environments and are comfortable working independently in a fully remote setting.