Prime Intellect
Research Engineer - Inference
$350K
San Francisco +1 · Full-time · Posted Oct 16, 2026
- Location
- San Francisco
Remote - Visa sponsorship
- Visa sponsorship is available for international candidates.
About the job
Build and optimize inference systems for Prime Intellect’s reinforcement-learning research, supporting large models and frequent model updates.
What you'll do
- Extend prime-rl’s vLLM integration to support new model architectures and validate them for RL training.
- Improve rollout throughput and optimize multi-node MoE inference, including concurrency, KV-cache reuse, and prefill/decode tradeoffs.
- Reduce trainer-to-inference update overhead; evaluate lower-precision inference and collaborate on inference-efficient model architectures.
What they ask for
- Experience building or optimizing GPU inference systems and understanding inference-engine internals.
- Strong Python skills and experience with PyTorch or a comparable ML framework.
- Ability to profile end-to-end workloads and verify 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.
Pay
$150-350k (currency not stated; cash compensation, including equity incentives)
Tools
prime-rl, vLLM, Python, PyTorch, GPU, MoE, FP8, CUDA
AI focus
Develop inference systems and optimize model execution for RL research, including large-scale models, long-horizon agents, and lower-precision inference.
Model-written summary · the original posting is authoritative