About the role

Research Engineer productionizing generative AI models for Waabi’s autonomous trucks and robotaxis. Optimizing GPU performance, simulation realism, and deployment tooling at Waabi.

Responsibilities

Own the path from research to production for neural and generative models powering Waabi World, a multi-sensor simulator
Partner with research teams on feed-forward scene reconstruction models, diffusion models, and vision foundation models
Turn research code and checkpoints into versioned, reproducible, exportable production models
Run production models in closed-loop simulation alongside GPU renderers and the autonomy stack
Support pipelines that reconstruct 4D simulation scenes
Profile model graphs, GPU kernels, and memory movement end to end
Apply model compilation, mixed precision, quantization, batching, CUDA graphs, custom kernels, faster sampling techniques, and distillation to meet latency and memory budgets
Build benchmarks and numerical-parity tests focused on simulation realism and downstream autonomy metrics
Catch performance and correctness regressions in CI
Design export, versioning, model registry, and build tooling for deploying models to simulation
Work with research scientists to make new architectures exportable, efficient, and hardware-friendly

Experience reading and developing production-quality software at scale, not only prototypes
Experience shipping ML systems that other engineers and researchers depend on every day
Experience taking models from a research codebase to an optimized production artifact and measuring both speedup and accuracy change
Hands-on experience with several of: TensorRT, torch.compile, torch.export, ONNX, post-training quantization, quantization-aware training, and pruning
Experience handling models with custom CUDA operators or other operations that do not export cleanly
Experience profiling with Nsight Systems, Nsight Compute, and the PyTorch profiler
Ability to reason about memory bandwidth, kernel launch overhead, and host-device synchronization
Ability to write or modify CUDA or Triton kernels
Ability to read research code for vision transformers, diffusion and flow-matching models, and vision foundation models
Knowledge of accuracy trade-offs when lowering precision, compiling a model, or reducing sampling steps
Experience launching and debugging GPU jobs in the cloud (AWS, GCP, or equivalent)
Experience with job orchestration and containerized environments
Ability to keep CUDA, PyTorch, and TensorRT versions compatible across a fleet

Competitive compensation and equity awards
Health and Wellness benefits encompassing Medical, Dental and Vision coverage (for full-time employees only)
Unlimited Vacation
Flexible hours and Work from Home support
Daily drinks, snacks and catered meals (when in office)
Regularly scheduled team building activities and social events both on-site, off-site & virtually