Senior Inference Engineer, AGI

Highlights

• Own inference for real-time multimodal conversational AI end to end.
• Make frontier speech and audio models feel human in real time at scale.
• Close the loop between research, production serving, and reinforcement learning.

Ideal candidate

A senior systems-minded engineer who thrives at the research-to-production boundary and obsesses over latency, cost, and human-feel interaction. You enjoy owning complex systems end to end and partnering with scientists to make ambitious models real.

Requirements

• Partner with research scientists to co-design servable model architectures
• Implement and optimize inference path for large-scale multimodal models including attention and KV-cache
• Apply efficiency techniques such as quantization, speculative decoding, operator fusion and paged KV-cache
• Develop high-performance kernels and profile with Nsight Compute/Systems and roofline analysis
• Own real-time streaming serving path meeting sub-second latency under concurrent load
• Build continuous batching, scheduling and preemption balancing throughput and latency SLAs
• Customize serving frameworks such as vLLM and PyTorch for real-time streaming models
• Implement multi-GPU inference with tensor parallelism and collective communication
• Build offline inference for post-training including rollout generation and reward-model serving for RL
• Ensure train/serve consistency across sampling and logit processing