Machine Learning Engineer

About Us

Shelfmark is the quality control platform that lets manufacturers catch defects the moment they happen. Our systems pair high-resolution line scan cameras with embedded software and machine learning to inspect products as they move down the line - flagging flaws in real time, at production speed, so our customers ship quality with confidence instead of relying on slow, manual spot checks.

As a Machine Learning Engineer, you'll own the models at the heart of that inspection pipeline. You'll work hands-on with imagery captured directly off the line, building and tuning the computer vision and anomaly-detection models that decide what's good and what isn't - then getting them running fast and reliably on the edge hardware that sits next to the camera. You'll work closely with our hardware and embedded engineers to ensure our models hold up against real world conditions.

What You'll Do

- Build, train, and evaluate ML models for defect detection and classification on imagery captured from live inspection lines.

- Develop ML models for anomaly detection and classification in cases where labeled defects are rare or hard to define.

- Optimize and deploy models to run on edge hardware at the inspection line, balancing accuracy against latency and throughput constraints.

- Partner with hardware and embedded systems engineers to integrate models into the end-to-end inspection pipeline, from camera capture to real-time decision.

- Establish and improve the data workflow - labeling, dataset curation, augmentation, and retraining loops - to keep models sharp as products and conditions change.

- Diagnose model performance issues in production and on-site, using real line data to drive improvements.

- Help define ML standards, tooling, and best practices that the broader team will build on.

What We’re Looking For

- Experience training, evaluating, and deploying deep learning models using Pytorch or Tensorflow, preferably for computer vision applications.

- Hands-on experience with classical image processing, unsupervised/self-supervised learning methods, preferably applied to anomaly detection. Solid grounding in classical computer vision and statistical modeling.

- Experience with or strong interest in vision-language models for tasks like zero/few-shot classification, and prompt driven anomaly detection.

- Familiarity with model optimization for edge deployments – quantization, ONNX/TensorRT, and profiling models under latency and memory constraints.

- A pragmatic, results-oriented mindset - comfortable working with messy real-world data and iterating quickly.

- Willingness to work on-site and collaborate closely with hardware and embedded teammates.

- Bonus: experience in manufacturing, industrial inspection, or other real-time/high-throughput vision systems.