Bolsa de trabajo en Argentina

We are looking for an AI Systems Engineer (CUDA/ONNX) based in Latin America Responsibilities Design, build, and maintain low-latency backend services and pipelines supporting real-time AI-powered multi-camera shading and visual matching. Integrate and deploy computer vision, scene analysis, and color-matching models using ONNX and ONNX Runtime within real-time inference pipelines. Optimize AI inference and image-processing workloads for NVIDIA GPUs using CUDA, TensorRT, or DeepStream, focusing on processing latency, throughput, and memory utilization. Develop multi-stream video processing pipelines capable of processing a minimum of 4 simultaneous camera feeds in real time, with architectural support to scale linearly to 12 feeds. Build systems for continuous scene analysis and visual matching across heterogeneous sources (live cameras, replay servers, and recorded clips). Implement algorithmic and AI-driven camera shading logic to calculate precise adjustments for Iris (exposure), Black level (pedestal), White balance (temp/tint), Gain, ND filters, and image detail/flare parameters. Develop camera control integration modules using broadcast protocols to send parameter corrections back to physical camera hardware. Support multi-mode shading execution within the engine, including fully automatic matching, operator-assisted suggestions, and manual parameter overrides. Optimize CPU-to-GPU memory transfer and data movement across multi-camera video decoding and model inference operations. Build robust APIs and WebSockets/gRPC services to expose real-time shading values, AI suggestions, and engine status to the web UI. Profile, troubleshoot, and resolve performance bottlenecks across GPU execution, video processing pipelines, and system memory. Implement monitoring and observability for frame drops, inference latency, GPU memory utilization, camera control command feedback, and overall system reliability. Collaborate with frontend, video, and product engineers to move real-time camera shading algorithms from prototype to production. Requirements Advanced level of English Strong professional experience in backend software engineering, using C++, Python, Go, or similar languages. Experience building production-grade APIs, microservices, or low-latency backend systems. Hands-on experience deploying AI/ML inference workloads in real-time or high-throughput production environments. Practical experience using ONNX and ONNX Runtime for model execution. Direct experience writing or optimizing code using CUDA or CUDA-enabled frameworks on NVIDIA GPUs. Solid understanding of camera shading concepts (exposure, pedestal/black level, gain,white balance, tint, color matching). Experience building multi-stream, low-latency video pipelines (handling multiple parallel video streams simultaneously). Understanding of digital video fundamentals: color spaces (Rec.

709, Rec.

2020, Log formats), frame rates, encoding/decoding, and uncompressed stream processing. Experience integrating hardware control interfaces or device APIs (e.g., IP camera Experience integrating hardware control interfaces or device APIs (e.g., IP camera protocols, serial/VISCA controls, or custom hardware endpoints). Strong understanding of concurrency, asynchronous execution, low-latency data structures, and memory management. Proficiency in profiling and troubleshooting CPU/GPU performance bottlenecks Bonus Points Bachelor’s Degree in Computer Science, Systems Engineering or related fields. Intermediate Level of Japanese. Experience with NVIDIA TensorRT, DeepStream SDK, NVENC/NVDEC, or Triton Inference Server. Experience with video processing and streaming tools like OpenCV, FFmpeg, GStreamer, NDI, RTSP, or WebRTC. Practical experience with color science calculations, vectorscopes, histograms, or custom CUDA image-processing kernels. Experience converting and optimizing models from PyTorch to ONNX/TensorRT for production. Knowledge of professional broadcast engineering, CCU/RCP workflows, or live multi-camera production control. Experience deploying GPU workloads using Docker and Kubernetes