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 on consumer-grade devices connected via the internet. The company recently achieved a major milestone with Agora, a permissionless run that pretrained an 8B model from scratch on consumer GPUs distributed across the internet, with no single participant ever holding the full weights.
As Research Scientist - Post-Training, you will own the end-to-end post-training stack that makes decentralized models useful for agentic applications. This is a hands-on technical role where you will both build systems and invent algorithms for a fundamentally new problem space.
Key responsibilities include:
- Building the complete RL training loop: rollout ingestion from a geo-distributed inference pipeline, reward computation, policy updates, and weight distribution back to the network. You set direction and drive execution.
- Inventing novel algorithms for asynchronous, high-latency, partially-trusted generation environments. Standard RL recipes assume on-policy rollouts from fast, trusted hardware; you will adapt them to decentralized constraints through staleness tolerance, off-policy corrections, and communication-efficient policy updates.
- Shipping the first post-trained decentralized models: building evaluation frameworks that demonstrate model improvement and taking the first decentralized post-trained release from experimental run to public artifact.
The role is deeply technical and research-oriented. You will work across timezones with a globally distributed team, primarily in Australia and North America. The company is backed by Union Square Ventures and other tier-1 investors, and the team consists of world-class ML researchers tackling open problems with no published solutions yet.
REQUIREMENTS:
- Hands-on experience running RL post-training on large language models (RLHF, RLVR, or reasoning-focused RL), with direct involvement in systems-layer work: rollout generation, async training loops, weight synchronization. Not launching jobs on someone else's stack.
- Strong production-level Python and PyTorch engineering: concurrency, failure handling, profiling and optimization.
- Research ability demonstrated through publications in RL post-training, asynchronous or distributed RL, or adjacent fields. Unpublished work you can defend in detail is also valued.
- Mission alignment: belief that Protocol Learning is a viable third path for collective, trustless, and sovereign AI.
- Professional-level English proficiency (written and spoken).
NICE TO HAVE:
- Experience training over slow networks or with decentralized/federated setups.
- Familiarity with serving-engine internals (vLLM, SGLang).
- Experience with reward modeling or verifiable-reward datasets.
- Experience with P2P networking and NAT traversal.
- Background at proprietary, open-weight, or open-source AI labs.
