Senior Software Engineer, Real-Time AI and Rendering - Holoscan SDK
NVIDIA
At NVIDIA, we are building the future of real-time AI for sensor-driven applications. Holoscan Platform is our flagship open-source framework for sensor AI, enabling developers to build, optimize, and deploy GPU-accelerated pipelines that process multimodal sensor data in real time. Originally pioneered for healthcare imaging, Holoscan has rapidly evolved into a domain-agnostic platform powering diverse industries-from humanoid robotics and industrial automation to astronomy, earth observation, and beyond. Wherever high-throughput sensor processing, AI inference, and visualization meet, Holoscan provides the foundation. The field of streaming AI and accelerated sensor processing is reshaping how humans interact with machines and how machines perceive the world. By combining advanced GPU computing, Vulkan-based rendering, and flexible software frameworks, Holoscan makes it possible to transform raw, high-bandwidth sensor streams into actionable intelligence and immersive visualization-spanning surgery, robotics, manufacturing, and scientific discovery.
Generative AI is becoming a central force in real-time sensing, simulation, and robotics - and this role will play a pivotal part in bringing those capabilities into Holoscan. Emerging multimodal foundation models (VLMs, video-language models, neural fields) are becoming essential for real-time perception. This role will extend Holoscan’s core mission by enabling GPU-resident generative methods that accelerate development, improve simulation fidelity, and unlock new possibilities for real-time perception. As part of this groundbreaking Holoscan SDK team is at the forefront of this transformation! We are designing APIs, building GPU-accelerated software pipelines, and collaborating with research and industry partners to deliver the foundation for tomorrow’s real-time AI platforms.
What You’ll Be Doing:
Architect the next generation of Holoscan SDK by developing intuitive, scalable APIs for real-time sensor, imaging, and multimodal data processing—balancing developer usability with peak GPU performance.
Prototype GPU-accelerated algorithms for computer vision, imaging, sensor fusion, and low-latency rendering-translating research into production-grade software.
Build and optimize core GPU libraries for accelerated I/O, streaming, decoding, and visualization, employing CUDA, Vulkan, and GPU-resident data paths.
Contribute to real-time visualization frameworks for medical, robotic, or industrial applications-integrating Vulkan, OpenGL, or Omniverse/RTX-based rendering back-ends.
Benchmark performance rigorously, profiling and optimizing across the full pipeline (Sensor → AI → Render → Display, Sensor → AI → Robotic Control).
Combine generative models with the Holoscan Sensor Bridge (HSB), Isaac Sim, ISAAC Lab and Omniverse to create real-time “AI-powered virtual sensors” that behave like real hardware—enabling development and testing long before physical sensors exist.
Prototype and optimize neural field (NeRF/SDF/Gaussian) operators for real-time scene reconstruction, view synthesis, and 3D perception—directly within Holoscan’s streaming architecture.
Integrate and optimize vision-language and multimodal foundation models for real-time, GPU-resident sensor pipelines
What we need to see:
A strong communicator and collaborator able to work across multiple domains-from AI and compute to graphics and visualization.
Dynamic programming expertise in C++ (modern standards), plus proven Python skills for prototyping and tooling.
Deep passion for real-time AI, computer vision, and sensor-driven systems-plus a passion for high-performance visualization and rendering.
Familiarity with multimodal or vision-language models and an understanding of how to adapt them to streaming or real-time workloads is a strong plus.
Success designing APIs and frameworks that stand the test of scale and that developers love to use!
8+ years of experience building and shipping complex, high-performance imaging, sensor, or rendering software.
Familiarity with GPU processing and rendering pipelines, synchronization, GPU memory management, and multi-GPU rendering is a plus.
Master’s/PhD or equivalent experience in Computer Science, Applied Math, Electrical or Computer Engineering, or related fields.
Ways to Stand Out from the Crowd:
Experience adapting VLMs or multimodal foundation models to real-time sensor or video pipelines.
Background integrating real-time GPU-accelerated processing and visualization pipelines (e.g., CUDA ↔ Vulkan interop),
Hands-on expertise with CUDA C/C++ and deep knowledge of GPU architecture and parallel programming paradigms.
Knowledge of Omniverse Kit, or other GPU rendering frameworks for real-time visualization.
Acute understanding of low-latency streaming pipelines for multimodal sensor fusion
You will also be eligible for equity and benefits.