Computer Vision Engineer

RealSense delivers industry-leading depth cameras and vision technology used in autonomous mobile robots, humanoids, access control, industrial automation, healthcare and more. With a mission to deliver world class perception systems for Physical AI and safely integrate robotics and AI into everyday life, RealSense provides

intelligent, secure and reliable vision systems that help machines navigate and interact with the human world.

About The Role

Facial Authentication is one of RealSense's flagship products - a secure biometric solution used in access control and identity verification around the world. It's a category of its own: high-accuracy, anti-spoofing face recognition designed for real-world security.

We are looking for a hands-on software engineer with deep expertise in machine learning inference and on-device optimization to join our team. You will play a central role in developing and maintaining the ML pipeline at the heart of our facial biometrics’ solution - running complex deep learning models efficiently on resource-constrained embedded hardware.

This is not a research role. We are looking for someone who gets things working in the real world - who can take a trained model, understand its computational demands, and engineer it to run fast, lean, and reliably on target hardware. You'll work hands-on in C++ and C on Linux-based embedded platforms, collaborating closely with algorithm and hardware teams to squeeze the best performance out of every deployment.

As part of a fast-paced startup, you'll be expected to take end-to-end ownership, make pragmatic engineering decisions, and contribute meaningfully to a product used in real-world applications.

What you’ll do

• Integrate, optimize, and maintain deep learning inference pipelines on embedded Linux platforms
• Optimize models for power, memory, and latency constraints - including quantization (INT8/FP16), pruning, and architecture-level trade-offs
• Work with inference frameworks including OpenCV, TensorFlow/TFLite, ONNX Runtime, and OpenVINO
• Profile and benchmark model performance on target hardware; identify and resolve bottlenecks
• Collaborate with algorithm teams to adapt and fine-tune models for deployment - understanding training outputs and translating them into production-ready inference
• Develop and maintain C++/C software components for real-time computer vision and biometric pipelines
• Ensure system stability, reliability, and reproducibility across hardware configurations
• Contribute to architectural decisions and code reviews with a focus on performance and maintainability
• Take end-to-end ownership of features from integration through deployment
  
  

• 4+ years of hands-on software engineering experience in C++ and C on Linux
• Proven experience optimizing deep learning models for resource-constrained environments - quantization, INT8/FP16 conversion, memory footprint reduction
• Hands-on experience with at least two of: OpenCV, TensorFlow / TFLite, ONNX Runtime, OpenVINO
• Familiarity with hardware-specific acceleration (NPUs, DSPs, GPUs) and associated SDKs
• Solid understanding of model training pipelines - not as a trainer, but enough to work effectively with algorithm teams and understand model outputs
• Strong profiling and debugging skills - comfortable with performance analysis tools on embedded Linux
• Deep understanding of memory management, compute constraints, and power budgets on embedded hardware
• Proven ability to work independently, own complex components, and deliver in a fast-moving environment
  
  

• Experience with computer vision tasks such as face detection, recognition, or liveness detection
• Experience with device–host communication protocols (e.g., USB, UART, TCP/IP)
• Background in signal processing or image processing pipelines
• Exposure to startup environments or fast-moving product development cycles