Senior RL Post-Training Engineer (Protocol Learning)

Pluralis Research is pioneering Protocol Learning, a decentralized approach to training and serving large models on consumer-grade devices connected via the internet. Despite initial skepticism, they’ve achieved breakthroughs, including Agora, a permissionless run that pretrained an 8B model from scratch using consumer GPUs without any single participant holding full weights. Explore their mission in full via A Third Path: Protocol Learning.

Post-training is critical for making these models useful for agentic use-cases. However, traditional post-training stacks rely on centralized infrastructure—synchronous rollouts, fast interconnects, and trusted workers. Pluralis’ approach operates entirely differently: it runs on consumer GPUs and Macs across the public internet, with no single participant holding the full model weights. Rollouts arrive from a geo-distributed inference pipeline with high latencies. Your role will be to make RL post-training work in this decentralized, high-latency environment—designing both the algorithms and the system end-to-end.

Key Responsibilities

• Build the post-training stack: Develop the RL training loop from start to finish, including rollout ingestion from the geo-distributed inference pipeline, reward computation, policy updates, and distributing updated weights back to the network. Set the direction and execute.

• Invent the algorithms: Adapt standard RL recipes to handle asynchronous, high-latency, and partially trusted generation scenarios. Focus on staleness tolerance, off-policy corrections, and communication-efficient policy updates.

• Ship first post-trained models: Create evaluation frameworks to demonstrate model improvements and lead the first decentralized post-trained model release from research to public artifact.

What We're Looking For

• Hands-on RL post-training experience: Direct experience with RL post-training on large language models, including RLHF, RLVR, or reasoning-focused RL. You must have worked with rollout generation, asynchronous training loops, and weight synchronization—not just launched jobs on someone else’s stack.

• Strong engineering skills: Proficiency in production-quality Python and PyTorch, with expertise in concurrency, failure handling, and profiling before optimization.

• Research ability: Publications or unpublished work in RL post-training, asynchronous or distributed RL, or related fields. You should be able to defend your work in detail.

• Mission alignment: Belief that Protocol Learning represents a viable third path for collective, trustless, and sovereign AI.

Nice to Have

• Experience training over slow networks or with decentralized/federated setups.

• Familiarity with serving-engine internals like vLLM or SGLang—the rollout pipeline is essentially a serving system.

• Experience with reward modeling or building verifiable-reward datasets.

• Experience with P2P networking and NAT traversal.

• Background at proprietary, open-weight, or open-source AI labs.

Compensation & Benefits

• Equity-Heavy Package: Significant ownership for key technical contributors, in addition to a high base salary.

• Remote-First Culture: Flexible work environment with globally distributed team members.

• Visa Sponsorship: Optional full visa sponsorship and relocation support to either Australia or the US.

• Open Problems: Training and serving frontier models on hardware you don’t control, over networks you don’t own, is largely uncharted territory. You’ll help write some of the first solutions.

FYI's

• The team works remotely across the world, primarily in Australia and North America. You’ll need to be comfortable collaborating across timezones.

• Professional-level English proficiency (written and spoken) is required.

• Recruiters: Pluralis is not currently seeking agency support. They’ll reach out if assistance is needed.