Senior Software Engineer (Embedded Vision Systems)

Vivid Machines; Toronto; On-site; CAD 120,000 - 180,000; Full time

Job summary

Vivid Machines is seeking a Senior Engineer to tackle challenges across embedded, cloud, and mobile software stacks. You will design and implement robust, real-time vision pipelines spanning cameras, sensors, and edge devices.

You will own the capture and inference pipeline, optimize performance under fixed power and thermal budgets, bring up hardware, and collaborate with a multidisciplinary team to ship scalable solutions for field deployments.

Qualifications

• Experience with cameras and sensors at a low level (image sensors, capture drivers).
• Strong background in embedded and real-time software for edge devices.
• Proficiency with C/C++ and Rust in systems programming.

Responsibilities

• Own the capture and inference pipeline for multi-camera video from sensor to model to storage.
• Optimize vision models to run in real time on constrained hardware with fixed power/thermal budgets.
• Bring up cameras and sensors at the driver level.
• Turn sensor data into geometry and fusing vision with position and motion data.
• Collaborate with a multidisciplinary team to deliver scalable field-ready solutions.

Skills

C/C++

Rust

Embedded systems

Real-time systems

Vision pipelines

Tools

ROS

GStreamer

Job description

• As a Senior Engineer, you will have the opportunity to work on challenging problems throughout our software stack, designing and developing robust and scalable solutions across embedded systems, cloud, and mobile
• You’ll be collaborating with our multidisciplinary team to create an amazing product and solve one of the world’s most important problems
• Do you love solving complex problems, and want to see your work applied in real life?
• We are looking for people who are self-directed and driven by a desire for excellence as much as by curiosity and a desire to learn by solving previously unsolvable problems
• Own the capture and inference pipeline. Multi-camera video from sensor to model to storage. No copies, no dropped frames, and it stays that way as resolution and model complexity go up
• Make heavy vision models run in real time on hardware you can’t upgrade. Fixed power and thermal budget
• Bring up cameras and sensors at the driver level
• Turn sensor data into geometry. Fuse vision, position and motion into per-tree geolocation you can defend against surveyed ground truth
• Make field problems reproducible at a desk. Replay of real scans, synthetic sources, on-device CI, regression tests against ground truth. Verify the fix before it ships
• Identify problems before our customers do: frame drops, latency, thermal and power headroom, sensor health, storage. The camera must notice its own problems
• Take our next generation platform from hardware definition to shipping product, help decide what that hardware should be
• Ship to a remote fleet that’s online intermittently

Experience with cameras and sensors at a low level. Image sensors, capture drivers, timing and synchronization, I2C/SPI/GPIO peripheralsExperience making a distributed or embedded system observable: metrics and logs from devices you can’t reach, dashboards and reports someone other than the author will use, and accuracy or quality tracked over time rather than measured onceComfort with Linux as an embedded platform: kernel customization and debugging, containers, cross-compilationHands-on experience deploying vision models to edge devicesStrong systems programming ability in a compiled language, such as C, C++, or rust, and real comfort at the boundary between application code, drivers and hardwareSelf-direction. This is a small, distributed team; the person in this role will often be the one who decides what “done” meansExperience building real-time streaming media or vision pipelines. Buffer management, latency and throughput, backpressure, zero-copy memory, and the discipline to profile instead of guessStreaming or event-based frameworks, such as ROS or gstreamer, including writing custom elementsAnd corresponding profiling and performance optimization techniquesEmbedded vision processors and neural network acceleratorsKalman filtering, multi-object tracking, and geometric state estimation or photogrammetryTelemetry and observability pipelines, time-series metrics, and building dashboards or automated reporting on top of themAgriculture, robotics, autonomy, or any other domain where the physical world refuses to cooperate