Computer Vision Engineer-NVIDIA Jetson (Edge AI/DeepStream/TensorRT) Mannai Corporation Q.P.S.C.

About the role

Ability to design, deploy, and optimize real time computer vision solutions on NVIDIA Jetson (Orin/Xavier/Nano series). Own the end to end pipeline, from camera ingestion and GStreamer/DeepStream graphs to model optimization with TensorRT and production deployment on JetPack Linux. Expect hands on work with CUDA, TensorRT, DeepStream, and Jetson performance tuning (nvpmodel/jetson_clocks), delivering robust, low latency analytics at the edge.

Key Responsibilities

• Design & implement CV pipelines on Jetson using DeepStream (GStreamer) for multi stream video analytics, integrating detectors/trackers/encoders and custom plugins in C++/Python.

• Optimize DNN models for edge inference with TensorRT (INT8/FP16 calibration, layer fusion, engine build, Torch TensorRT when applicable).

• Model selection & deployment (YOLO/DetectNet/SegNet, etc.) with performance profiling and accuracy trade offs specific to Jetson hardware capabilities.

• Camera & I/O integration: CSI/USB/IP cameras; build robust GStreamer pipelines, configure encoders/decoders (H.264/H.265/AV1), and manage RTSP ingestion.

• Power/performance tuning (nvpmodel, jetson_clocks) to meet latency/thermal/power budgets across Orin NX/Nano modes (10–25W).

• JetPack environment management: flash/upgrade via SDK Manager or Debian packages; containerize workloads; maintain reproducible build/deploy processes.

• Production hardening: logging/metrics, OTA update strategy, graceful failure/restart, device health monitoring.

• Collaborate with data engineers & product teams on accuracy/performance SLAs, test plans, and on device evaluation.

Required Skills & Experience (must have)

• 3–7 years in computer vision or embedded AI, with hands on Jetson deployments (Orin/Xavier/Nano).

• Strong Python/C++ for performance critical code (NumPy/CuPy/OpenCV/GStreamer bindings).

• DeepStream proficiency (graph design, deepstream-app configs, custom plugins).

• TensorRT experience: building engines from ONNX/PyTorch, calibration to INT8/FP16, profiling, and troubleshooting.

• Practical CUDA knowledge (streams, memory, pinned buffers) and JetPack/Linux device operations (L4T, Ubuntu, drivers).

• Camera stack familiarity (CSI/USB/IP), media codecs (H.264/H.265/AV1), and GStreamer basics.

• Performance tuning on Orin NX/Nano (nvpmodel/jetson_clocks), thermal and power constraints.

Nice to Have (preferred)

• Experience with Ultralytics YOLO on Jetson via DeepStream/TensorRT; knowledge of TAO Toolkit.

• Knowledge of VPI (Vision Programming Interface) for accelerated CV primitives.

• Containerization on Jetson (Docker/OCI), Triton Inference Server integration on edge.

• Sensor fusion, multi camera synchronization, time stamping.

• Deployment in industrial environments (latency/uptime/security constraints).

• Experience with Torch TensorRT on JetPack 6.

Education

BSc/MSc in Computer Science, Electrical/Computer Engineering, or equivalent practical experience.