Computer Vision Engineer

• Mosaic’s team is seeking a Computer Vision Engineer to take vision models from development into reliable, real-time production on the edge.
• The consultant will optimize and deploy models (PyTorch to ONNX/TensorRT) on NVIDIA Jetson Orin embedded hardware among others, running in containerized ARM64 environments, to support process control, safety, and quality use cases across mining and chemical plant operations.
• Representative use cases include real-time visual monitoring, equipment-state detection, anomaly identification, and other camera-based applications that support operational decision-making and integration with downstream systems.
• The role is hands-on and delivery-focused: performance, latency, and long-run stability on the device matter as much as model accuracy.

Focus area

• Edge model optimization & deployment 40%
• Real-time video pipeline engineering 35%
• Validation, profiling & production support 25%

Responsibilities

Edge Model Optimization & Deployment (~40%)

• Convert and optimize PyTorch vision models (detection, segmentation, classification) to ONNX and TensorRT engines for NVIDIA Jetson Orin.
• Package inference workloads as Docker containers on JetPack/L4T (ARM64) and manage reproducible builds and deployments to field devices.
• Partner with data scientists to make models edge-ready (input sizing, architecture trade-offs, pre/post-processing).

Real-Time Video Pipeline Engineering (~35%)

• Build and tune low-latency RTSP/GStreamer ingestion pipelines using hardware decode and zero-copy (NVMM) memory.
• Develop DeepStream pipelines, including custom C++ parsers/plugins for non-standard model outputs where required.
• Write performance-critical components in C++ and CUDA; use Python for integration, tooling, and prototyping.
• Deliver model outputs to downstream systems (e.g., alerts, dashboards, PLC/control-system interfaces) in coordination with site automation teams.

Validation, Profiling & Production Support (~25%)

• Profile end-to-end latency, throughput, GPU/CPU/memory utilization, and thermals (e.g., Nsight Systems, tegrastats) and remove bottlenecks.
• Engineer for 24/7 stability: stream reconnection, watchdogs, backpressure and frame-drop control, logging, and health monitoring.
• Support site commissioning, field testing, and troubleshooting, including onsite visits to mining and plant locations as needed.
• Document deployment procedures, configurations, and runbooks so solutions can be reused and scaled across sites.

• Mid-to-senior level experience in computer vision / ML engineering with proven delivery of models to embedded or edge vision systems in production.
• Hands-on experience with NVIDIA Jetson Orin (AGX Orin, Orin NX, or Orin Nano).
• Experience optimizing and deploying models with TensorRT and ONNX.
• Proficiency in Python and PyTorch.
• Docker containerization on ARM64 / JetPack / L4T.
• Experience profiling and tuning for latency and long-running stability on constrained hardware.
• Strong written and verbal communication; able to work directly with plant operations, automation, and IT/OT teams.

Preferred Qualifications

• Strong NVIDIA DeepStream SDK experience, including custom plugins or output parsers.
• Strong CUDA and C++ development skills.
• RTSP/GStreamer video pipeline development with IP cameras.
• Experience in industrial environments (mining, chemicals, manufacturing, energy) and with OT/control-system integration (PLC, OPC UA, Modbus).
• Familiarity with ruggedized camera and edge hardware selection, network and power constraints at remote sites.
• Experience with fleet management, remote updates, or monitoring for multiple edge devices.
• Ability to work within enterprise change-control (MOC) and cybersecurity review processes.