Member Of Technical Staff - Research Engineer (AI)

TL;DR

Member of Technical Staff - Research Engineer (AI): Developing and optimizing large-scale training systems for multimodal generative models with an accent on GPU performance, numerical stability, and distributed training. Focus on implementing custom kernels, low-precision training paths, and debugging complex distributed training failures to enable frontier research.

Location: San Francisco (USA) or Freiburg (Germany). Hybrid (at least 2 days a week) or remote with a required monthly in-person week.

Salary: $180,000 - $290,000 + equity

Company

A frontier research lab behind foundational technologies like Stable Diffusion and FLUX, creating advanced generative models for images and video.

What you will do

• Optimize the performance, reliability, and numerical stability of production training runs for large multimodal generative models.
• Profile full training steps across model code, attention, kernels, data loading, and communication.
• Implement GPU-level optimizations using CUDA, Triton, CuTe, and CUTLASS.
• Develop and validate low-precision training paths including FP8, MXFP8, and FP4-style formats.
• Debug distributed training failures such as NaNs, loss spikes, and NCCL issues.
• Build benchmarking and profiling harnesses to validate performance across various hardware and configurations.

Requirements

• Deep experience with large-scale training systems and strong PyTorch fluency.
• Proficiency in distributed training concepts (FSDP, tensor/model parallelism, NCCL).
• Hands-on experience improving training throughput, memory footprint, or stability.
• Experience profiling GPU workloads with Nsight Systems, Nsight Compute, or torch profiler.
• Understanding of low-precision training and quantization tradeoffs (FP8, FP4).
• Must be based in or able to travel to San Francisco or Freiburg for monthly in-person weeks.

Nice to have

• Experience co-owning training for a shipped frontier foundation model.
• Proven ability to write or substantially improve forward/backward GPU kernels.
• Experience with Hopper or Blackwell-class GPUs.
• Background in diffusion, flow matching, DiT, or LLM training systems.

Culture & Benefits

• Distributed team with physical offices in SF and Freiburg.
• Company covers reasonable travel costs for required in-person weeks.
• Culture based on scientific obsession, low ego, boldness, and kindness.
• Equity compensation provided alongside base salary.