Spatial World Model Researcher

We are hiring systems engineers, kernel hackers, world-model researchers, and roboticists to compile the next industrial paradigm. Layer 01: COMPUTE Remote / Global Full-time CUDA / Kernel Optimization Engineer Design and benchmark custom FlashAttention-3, TensorRT-LLM, and speculative decoding kernels targeting NVIDIA H100 clusters. Driving the marginal cost of continuous agentic reasoning toward zero. Required Skills: Deep CUDA / C++ / Triton expertise Experience with vLLM, TensorRT-LLM, or Megatron-LM Low-level GPU profiling (Nsight Systems) Develop multi-dimensional tensor manifolds [X, Y, Z, τ, σ] and spatial diffusion models capable of continuous geometric reasoning, physics simulation, and deterministic tool manipulation. Required Skills: PyTorch / JAX distributed training Expertise in 3D computer vision, NeRFs, 3D Gaussians, or world models Strong mathematical foundations in differential geometry Architect asynchronous goal-conditioned reasoning DAGs, persistent graph memory layers, and Model Context Protocol (MCP) tool execution engines for 100+ agent swarms. Required Skills: Distributed systems in Rust / Python / TypeScript Experience orchestrating complex multi-agent execution graphs Native understanding of MCP & tool calling reliability Build the volumetric Air Traffic Control cockpit in WebXR and native VisionOS to connect human supervisors with autonomous swarms using gaze vectors and sub-millimeter gesture recognition. Required Skills: WebXR / Three.js / WebGL / VisionOS Spatial UI/UX design for real-time telemetry rendering Low-latency WebSockets & gRPC data streaming Build the bridge between agent trajectories and physical robot actuation. Specialize in NVIDIA Isaac Sim domain randomization, kinematic solvers, and low-latency joint actuation loops. Required Skills: NVIDIA Isaac Sim / Omniverse / OpenUSD ROS2, C++, and robot kinematics (manipulators / AMRs / humanoids) Physical hardware deployment and telemetry streaming Don't see an exact title? If you have written custom kernels, trained world models, or built autonomous swarms, reach out directly. #J-18808-Ljbffr Reality 5D