Research Engineer: High-Performance ML Infrastructure

Fleet AI, Inc.; San Francisco (CA); On-site; USD 180,000 - 240,000; Full time

Job summary

Fleet AI, Inc. is seeking a Member of Technical Staff, Research Engineer, Infrastructure to own and optimize the large GPU training cluster and the end-to-end research stack in a fast-moving environment.

You will profile bottlenecks, implement kernel and system-level fixes, and collaborate with researchers to ensure reliable, scalable training and inference at scale onsite in San Francisco.

Qualifications

• Experience with multi-node training and major parallelism strategies (DP, TP, PP, FSDP).
• Ability to profile slow training runs and implement fixes in CUDA/Triton kernels or data pipelines.
• Experience building or maintaining high-throughput inference stacks (vLLM, SGLang) and KV cache tuning.
• Ability to navigate large, evolving research codebases and improve performance without breaking correctness.

Responsibilities

• Own the training cluster: capacity, failover, fault tolerance, on-call when training stalls.
• Profile training runs to identify bottlenecks (NCCL, KV cache, dataloader, allgather) and implement fixes; write kernels when needed.
• Maintain training and inference pipelines: distributed training (FSDP, TP, PP), rollout generation, async training, inference stack dependencies.
• Maintain the research codebase: keep it fast and reliable as researchers extend it without breaking the science.

Skills

Distributed training

Kernel optimization

Inference systems

Research-codebase maintenance

Tools

CUDA

Triton

Job description

Fleet AI, Inc. is seeking a Member of Technical Staff, Research Engineer, Infrastructure to own and optimize the large GPU training cluster and the end-to-end research stack in a fast-moving environment.

You will profile bottlenecks, implement kernel and system-level fixes, and collaborate with researchers to ensure reliable, scalable training and inference at scale onsite in San Francisco.