Research Engineer - Inference

The role

Build inference systems that power RL research for 1–3T parameter models on NVL72 clusters. Work involves improving rollout throughput and optimizing multi-node MoE inference.

What you will do

• Extend prime-rl’s vLLM integration to bring new model architectures into RL training
• Improve rollout throughput for long-horizon agents by tuning concurrency and KV-cache reuse
• Optimize multi-node MoE inference including expert parallelism
• Reduce trainer-to-inference weight-update overhead
• Build on FP8 and evaluate lower precision
• Use trainer/sampler mismatch checks to catch numerical regressions
• Work with model researchers on architecture choices to improve inference efficiency

Required qualifications

• Experience building or optimizing GPU inference systems
• Understanding of inference-engine internals
• Strong Python skills
• Experience with PyTorch or a comparable ML framework
• Ability to profile end-to-end workloads and turn measurements into verified performance improvements
• Experience debugging numerical or integration issues in ML systems

Nice to have

• Contributions to vLLM or another inference engine
• CUDA or Triton kernel work, particularly for low-precision inference

Benefits

• Cash Compensation Range of $150-350k, including equity incentives
• Flexible work arrangements
• Option to work remotely or in-person at San Francisco offices
• Visa sponsorship and relocation assistance for international candidates
• Quarterly team off-sites, hackathons, conferences and learning opportunities

Before you apply

• Visa sponsorship and relocation assistance available for international candidates

About the company

Prime Intellect is building the open superintelligence stack. The platform Lab unifies compute, environments, evaluations, secure sandboxes, high-performance training, and deployment.