SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Pluralis Research is building Protocol Learning: a system for training and serving large language models in a fully decentralized way across consumer-grade devices connected via the internet. The company has already demonstrated feasibility with Agora, a permissionless run that pretrained an 8B model from scratch on consumer GPUs distributed globally, where no single participant held the full model weights.
As a Research Scientist, you will identify and solve the open research problems that block Protocol Learning from scaling to frontier models. These include communication-efficient training across different parallelism axes, fault tolerance as nodes join and drop mid-run, heterogeneous compute and network conditions, and robustness to malicious participants. Your work will be published in top-tier venues (NeurIPS, ICML, ICLR) and directly integrated into live training runs by the engineering team.
Key responsibilities include: identifying the blocking questions for Protocol Learning at scale; publishing foundational papers in tier-1 conferences; and collaborating with the engineering team to ensure your methods land in production training runs, not just academic papers.
You should have a PhD in machine learning with publications in top-tier conferences, hands-on experience in large-scale distributed training and compression strategies, and strong PyTorch programming skills. Experience with foundation model pre-training, post-training, or RL is a plus, as is prior work at proprietary, open-weight, or open-source AI labs.
The company is backed by Union Square Ventures and operates a remote-first culture with teams distributed across Australia and North America. They offer an equity-heavy compensation package with significant ownership for key technical contributors, optional visa sponsorship and relocation support, and the opportunity to work on largely unsolved problems in distributed AI training.
