Member of Technical Staff - ML Performance
Veeda AI · Toronto · Zürich · Seattle · California
Roboticsremote
At a glance
Highlights
- Own throughput for multi-node video world model training from day one.
- Build next-gen multimodal world models for Physical AI with small fast team.
- Make interruptions cost minutes with resilient checkpointing and diagnostics.
Ideal candidate
A systems-minded performance engineer who loves chasing step time and stability on large-scale runs and wants outsized ownership building Physical AI with a fast-moving research team.
Requirements
- Bachelor's degree or equivalent hands-on experience in Computer Science, Computer Engineering, or related technical field.
- Deep hands-on experience with PyTorch and large-scale parallelism (FSDP2, Megatron-Core, TorchTitan, or DeepSpeed) on real multi-node jobs.
- Fluency in Python and C++/CUDA with ability to predict kernel stalls from memory access patterns.
- Experience profiling live training and inference runs and translating traces into step-time or MFU gains.
- Expertise in at least one of low-precision numerics, kernel authoring, or large-run fault diagnosis.
Lab profile
About Veeda AI
Building multimodal foundation world models for Physical AI.
What they build
- Multimodal foundation world models for Physical AI
Known for
- Multimodal foundation models
- World models
- Robotics
- Embodied intelligence
Key facts
- Small, fast-moving team
Posted · Last seen