Computer Vision Engineer

Also known as: CV Engineer, Computer Vision Developer, Computer Vision Scientist, Perception Engineer, Machine Vision Engineer, Vision AI Engineer.

What does a Computer Vision Engineer do?

A Computer Vision Engineer is an engineer who builds systems that recognise objects, people, text and events in images and video and makes them work in real conditions: in poor light, on a cheap camera, on a device without a powerful GPU. Typical tasks:

• build and label datasets, looking for skews and rare cases in them;
• train and fine-tune models for detection, segmentation, tracking and OCR, and increasingly adapt multimodal models;
• compress and speed up models for specific hardware: cameras, phones, edge devices, GPU servers;
• build the video processing pipeline from the camera to the result in the product;
• monitor quality after release and analyse the model's errors on live data.

When does a company need a Computer Vision Engineer?

A company needs a Computer Vision Engineer when its product depends on what a camera sees: quality control in manufacturing, retail and warehouses, medical imaging, robotics and autonomous vehicles, document verification, video analytics, AR. A common trigger is a prototype on an off-the-shelf API or a multimodal LLM hitting a limit in accuracy, latency or cost. Another is a product moving from the server to the device, where the model must fit constrained hardware. The role will not work if an off-the-shelf cloud recognition service covers the task and no in-house model is planned: then an AI / LLM Engineer or a Backend Developer with integration experience is enough.

Computer Vision Engineer vs ML Engineer: what's the difference?

A Computer Vision Engineer differs from an ML Engineer in depth in one modality: they specialise in images and video, while an ML Engineer works with any data — tables, text, events. Computer vision has its own difficulties that general ML experience does not cover: labelling, lighting and optics, video stream synchronisation, running on weak hardware. What sets them apart from a Data Scientist is responsibility for the model in production rather than for research.

Why is it hard to hire a Computer Vision Engineer?

Computer Vision Engineers are hard to hire because one title hides very different profiles, and experience in real conditions is much rarer than experience on training datasets.

• Researchers with publications, developers who connected an off-the-shelf API and embedded systems specialists all call themselves Computer Vision Engineers. Searching by title mixes them all together.
• Many candidates know computer vision from competitions and clean public datasets, and have never seen a model break on glare, rain or a new camera model.
• Edge experience — quantisation, compression, running on specific chips — is rare, and in a CV it is hard to tell apart from a line saying "familiar with TensorRT".

What to decide before hiring a Computer Vision Engineer?

• Where will the model run: in the cloud, on your own servers or on a device? Edge and server are different candidate pools, and getting this wrong screens out exactly the people you need.
• What is the task: detection, segmentation, tracking, OCR, 3D? Strong experience in document recognition does not mean a person can handle tracking people in video.
• Is there data, and who labels it? If there is no labelling, the Computer Vision Engineer will spend the first months building a data collection process, and the role must allow for that.
• An in-house model or an adapted multimodal one? The answer decides whether you need experience training from scratch or experience fine-tuning and evaluating existing models.

Where to find a Computer Vision Engineer?

Computer Vision Engineers most often come from robotics and autonomous vehicles, video analytics and smart camera companies, medtech working with medical images, fintech with document verification, and industrial automation. A strong signal is a model that ran on real cameras or devices and a story of which errors appeared after launch and how they were closed. NEWHR finds such candidates by the results of their work, not by applications to a job post.

How to assess a Computer Vision Engineer in an interview?

Computer Vision Engineers are assessed with a review of a system the candidate took to production and a technical interview with code or a case on the company's data. The review covers the whole chain: data and labelling, metrics, hardware, errors on the live stream. On the company side, an ML Lead or the CTO takes part, and if the model runs on a device, the engineer responsible for the hardware.

Strong signals:

• starts with the data: where it comes from, how it was labelled, which cases are missing;
• names specific errors the model made in real conditions and explains how they fixed them;
• knows how many frames per second their model delivers and on which hardware.

Red flags:

• all their experience is on public datasets and competitions;
• discusses only model architectures and cannot describe the pipeline as a whole;
• does not distinguish between a validation metric and quality for the user.

How long does it take to hire a Computer Vision Engineer?

We present the first candidates for the Computer Vision Engineer role within 5–7 days of the brief.

Related roles

• ML Engineer — may be enough if images are only one of several data types in the product.
• Neural Network Optimization Engineer — takes a vision model that already works and speeds it up on the target chip, while the Computer Vision Engineer trains it.
• AI Infrastructure / Inference Engineer — builds the GPU platform on which vision models process video streams in the cloud.
• Data Scientist — needed earlier, to check whether camera data answers the business question at all.