Computer Vision Engineer

Cruxx; New Jersey; On-site; USD 120,000 - 170,000; Full time

Job summary

Cruxx, a MedTech company in Central New Jersey, seeks an experienced AI/ML engineer to bridge research and product. You will take segmentation, registration and simulation methods from the lab and make them fast, robust, and deployable in the operating room, while informing the next model with real-world data.

You will split time between training/evaluating models, writing inference code, and building data/evaluation infrastructure to measure impact and reliability for regulatory-ready AI in a

Qualifications

• Three or more years applying deep learning to real computer-vision problems.
• Experience shipping a model from research to a shipped product or deployed system.
• Strong Python and PyTorch; working C++ for inference and integration.

Responsibilities

• Develop and productionise computer-vision and deep-learning models for segmentation, surface reconstruction and registration from live imagery.
• Optimise inference for real-time performance on GPU hardware: quantisation, TensorRT or equivalent, memory and latency budgets that hold under load.
• Build and maintain the data pipeline: capture, annotation, dataset versioning, training, evaluation and deployment.
• Design evaluation that reflects clinical reality: ground-truth protocols, failure-mode analysis, accuracy across patients, lighting and occlusion.
• Document models, data and decisions to the standard a regulator will expect for AI in a medical device.
• Collaborate closely with the navigation software team so models and pipeline are designed together, not bolted on.

Skills

Python

PyTorch

3D vision

Deep learning

Education

MSc in computer vision or ML

PhD in computer vision or ML

Tools

C++

TensorRT

Job description

A MedTech company, Central New Jersey. On-site. Reports to the Senior Software Engineering Leader, working day to day with the AI research team.

The problem

Trackerless registration means the system has to find the target, understand its surface and lock a pre-operative model to it, in a live surgical field, with blood, tools and hands in the way, to sub-millimetre accuracy and in under a second. Nobody has shipped this. Our client has it working in the lab and the numbers are getting better every month. The gap between a lab result and a cleared device is where you come in.

The role

You sit between research and product. The AI team builds segmentation models, registration methods and the simulation that turns pre-operative imaging and intra-operative data into a recommended plan. You take those methods and make them fast, robust and deployable on the compute in the operating room, then feed what you learn in the real world back into the next model. Expect to split your time between training and evaluating models, writing the inference code that runs them, and building the data and evaluation infrastructure that tells the team whether a change made the system better or worse.

What you will do

• Develop and productionise computer-vision and deep-learning models for segmentation, surface reconstruction and registration from live imagery
• Optimise inference for real-time performance on GPU hardware: quantisation, TensorRT or equivalent, memory and latency budgets that hold under load
• Build and maintain the data pipeline: capture, annotation, dataset versioning, training, evaluation and deployment
• Design evaluation that reflects clinical reality: ground-truth protocols, failure-mode analysis, accuracy across patients, lighting and occlusion
• Document models, data and decisions to the standard a regulator will expect for AI in a medical device
• Collaborate closely with the navigation software team so models and pipeline are designed together, not bolted on

What they are looking for

• Three or more years applying deep learning to real computer-vision problems, with at least one model you took from research to a shipped product or deployed system
• Strong Python and PyTorch; working C++ for inference and integration
• Solid grounding in 3D vision: multi-view geometry, depth estimation, point clouds, surface registration (ICP and learned variants), sensor calibration
• Experience with GPU inference optimisation and the tooling around it
• Rigorous about evaluation. You would rather report a worse number you trust than a better one you do not
• An MSc or PhD in computer vision, machine learning or a related field, or equivalent evidence of depth

Nice to have

Medical imaging (CT or MRI segmentation), surgical video or OR data, physics-based simulation, experience with AI in a regulated setting (FDA guidance on AI/ML-enabled devices, IEC 62304), publications or open-source work in 3D vision.

Who should not apply

If your interest stops at the leaderboard score, or you want to work only on models and never on the system they run in, this is not the role. Everything here has to work in a room with a patient on the table.