Neural AI
Embedded Vision Software Engineer
1
Which company are you aiming at?
To generate precise rejection-prevention feedback and application strategy, tell us where you're targeting.
Apple
NVIDIA
Tesla
Qualcomm
Bosch
Garmin
2
Send your application
You need an account so we know where to send the feedback.
The job description
Tech stack. C/C++, Python, OpenCV, TensorRT/ONNX Runtime, ARM Cortex-A + GPU/NPU, MIPI CSI camera pipelines, GStreamer, edge inference optimization
About the role
You will join a smart-camera company building edge AI products for retail analytics and industrial inspection. The embedded vision team owns the full pipeline from image sensor to inference result: camera drivers, image processing, and neural network deployment on the device. This role matters because the product promise is intelligence without the cloud, and that only works if the whole pipeline runs fast, cool, and accurately on a power-constrained SoC.
What you will achieve
- Ship an edge inference pipeline running object detection at 30 FPS on an ARM SoC with NPU, holding mAP within 2 points of the cloud model on the validation set, with accuracy reports generated for every model update
- Drive a 40% reduction in inference latency through quantization (INT8), operator fusion, and NPU-specific graph optimization, measured end to end on device
- Reduce camera pipeline latency from sensor to first processed frame below 100 milliseconds by tuning MIPI CSI, ISP, and DMA paths with frame-accurate measurement
- Build an automated model-deployment flow that takes a trained model to on-device validation in under a day, used by the ML team for every release candidate, and documented so new ML engineers onboard without help
- Own the vision accuracy regression suite, catching model or pipeline degradations before they reach customer pilots and blocking releases on accuracy drops
What you will bring
Must-haves
- 2-5 years building vision software on embedded targets, not just desktop OpenCV, with deployed product experience
- Strong C/C++ with image processing fundamentals: color spaces, filtering, and geometric transforms
- Experience deploying neural networks on edge accelerators such as NPUs, GPUs via TensorRT, or DSPs
- Hands-on camera pipeline work: MIPI CSI bring-up, ISP tuning, and V4L2 driver interaction
- Model optimization experience: quantization, pruning, or distillation applied for real deployment gains
- Python for prototyping, dataset tooling, and accuracy analysis that informs deployment decisions
- Performance profiling on heterogeneous SoCs: identifying whether the bottleneck is sensor, memory, or compute
Nice-to-haves
- Experience with specific edge platforms such as NVIDIA Jetson, NXP i.MX with NPU, or Qualcomm QCS
- Knowledge of video codecs (H.264/H.265) on embedded hardware encoders
- Familiarity with 3D vision such as stereo or ToF, or multi-camera synchronization