Principal Deep ISP AI/ML Engineer: Sensors
Israel
Posted 1d ago
Remote Work Policy
On-site
Categories
Machine Learning Engineer
About the job
What you'll do
Own the learned pipeline. Architecture, training and evaluation of a model that replaces the hand-tuned chain, starting with the camera ISP and low-light enhancement and growing to the full chain.
Make it run at the sensor. Design networks for Mythic's analog compute: quantization- and hardware-aware training, memory and bandwidth budgets, and the mapping work with our AI engineering team.
Extend beyond the camera. Take the same approach to other sensing modalities — radar, lidar and more — and to the fusion between them.
Build the quality science. Stand up the lab, define objective, perceptual and task-level metrics, and benchmark against best-in-class pipelines.
Build the data engine. Capture rigs, proprietary datasets, sensor-noise simulation, and a toolchain for per-sensor and per-application retraining and auto-tuning.
Deliver to customers. Demos on customer data and first design-ins across our target markets, working directly with business development and customer engineers.
Set the bar and grow the group. Hire and mentor the next engineers, publish and patent, and make Mythic a recognized name in learned sensing.
What you'll bring
10+ years in computer vision, imaging or sensor signal processing with deep learning, and still hands-on.
MSc or PhD in EE, CS or a related field.
Grounding in computational imaging: RAW-domain processing, noise, denoising, HDR or color.
Have taken models from research to a shipped product, owning data, training and evaluation.
Comfortable on constrained hardware: quantization, edge NPUs, compute and memory budgets.
Preferred: work on other sensing modalities — radar, lidar, depth or thermal — and multi-sensor fusion.
Preferred: publications at CVPR, ICCV or ECCV, or granted patents.
Preferred: non-standard or in-memory accelerators, or sensor-level work; led a small team or built a product from zero.
Ability to travel up to 10% of the time.
