Edge AI / Computer Vision Expert

We are looking for a strong Edge AI / Computer Vision Expert who can help us build, optimize, deploy, and improve real-time perception models running on edge devices inside hospital environments. About the Role This is a hands-on technical role for someone who can work across computer vision models, edge deployment, real-world sensor data, and production AI systems. You will work on human detection, room activity understanding, object/event detection, OR workflow recognition, safety monitoring, and multi-sensor perception. The ideal candidate can take models from experimentation to reliable deployment on edge hardware, with strong attention to accuracy, latency, false positives, and real-world robustness. Responsibilities * Develop and improve computer vision models for real-time room activity understanding. * Work on human detection, occupancy detection, object detection, event detection, and workflow phase recognition. * Optimize AI models for edge deployment on devices such as NVIDIA Jetson. * Convert, quantize, benchmark, and deploy models using tools such as TensorRT, ONNX, CUDA, PyTorch, or similar frameworks. * Improve model performance under real-world conditions such as occlusion, low light, reflections, camera angle changes, staff movement, and room layout variation. * Build and maintain evaluation pipelines for precision, recall, false positives, false negatives, latency, and edge resource usage. * Work with RGB, IR, depth/ToF, thermal, motion, and other sensor signals when applicable. * Design post-processing logic, tracking, temporal smoothing, and rule-based layers to improve reliability. * Collaborate with robotics, cloud, product, and QA teams to deploy models safely and reliably in hospitals. * Analyze field failures and improve models based on production data and edge logs. * Help define data collection, annotation, QA, and model validation workflows. * Support production deployment, monitoring, rollback, and continuous model improvement. What We’re Looking For * Strong experience in computer vision and deep learning. * Hands-on experience with object detection, human detection, segmentation, tracking, pose estimation, or video understanding. * Strong Python and PyTorch experience. * Experience deploying models on edge devices, especially NVIDIA Jetson or similar embedded GPU platforms. * Experience with ONNX, TensorRT, CUDA, OpenCV, GStreamer, or similar edge AI tools. * Ability to optimize for latency, memory, GPU utilization, power usage, and real-time performance. * Strong understanding of model evaluation, dataset quality, failure analysis, and production ML metrics. * Ability to work with noisy real-world data, not just clean benchmark datasets. * Comfortable debugging production issues across model behavior, camera input, hardware limitations, and edge runtime constraints. * Strong communication skills and ability to work closely with a fast-moving engineering team. Nice to Have * Experience with ROS2, robotics perception, or autonomous systems. * Experience with multi-camera or multi-sensor fusion. * Experience with video activity recognition or temporal event detection. * Experience with healthcare, operating rooms, patient rooms, or safety-critical environments. * Experience with YOLO, D-FINE, RT-DETR, SAM, ByteTrack, DeepSORT, MediaPipe, or similar models/tools. * Experience building data annotation workflows and active learning loops. * Experience with cloud-connected edge devices, OTA model deployment, telemetry, and monitoring. * Experience with privacy-preserving AI or on-device inference systems. Ideal Candidate The ideal candidate is a practical, production-minded computer vision engineer who can build models that work reliably in messy real-world environments. You should care not only about model accuracy, but also about latency, stability, edge performance, safety, and operational reliability. You should be able to move fast, test ideas quickly, analyze failures carefully, and turn field data into model improvements. You should also be comfortable working with both deep learning models and classical computer vision or rule-based logic when that is the right solution. Example Projects * Improve human detection reliability in hospital rooms with changing lighting, occlusion, and complex staff movement. * Optimize a detection model to run in real time on NVIDIA Jetson with low latency and stable GPU usage. * Build an evaluation pipeline for human detection and room activity recognition across different hospitals and room types. * Develop temporal logic to reduce false positives and false negatives in occupancy detection. * Detect OR workflow phases such as cleaning, setup, idle time, patient-in-room, procedure activity, and turnover completion. * Create model monitoring metrics for production edge deployments. * Convert and benchmark PyTorch models using ONNX and TensorRT. * Analyze field logs and videos to identify model failure modes and propose improvements.