30 - Computer Vision Engineer

About the role

We're looking for a computer vision engineer to join a small, autonomous team at an European startup building AI-powered machinery for automated visual inspection and sorting of physical materials. Their systems are already live in production at multiple client sites — this is not a research project, it's real machines making real-time decisions.

The software stack is in good shape overall. The gap is specifically on computer vision. You'll work alongside the current computer vision engineer on real-time detection of features and attributes on physical items moving through the system. Real-time performance is a hard requirement — this is not an offline or batch process.

The team is fast-paced and results-driven. You'll be expected to own your topic without close supervision.

What you'll work on

• Real-time object detection and segmentation models running on live production machinery

• Detection of fine-grained features and attributes on varied, irregular physical items

• Model optimization for real-time inference performance

• The full model lifecycle: training, registry, deployment, and monitoring

Must-have skills

CV models

• RF-DETR

• YOLO segmentation

• Sliced/SAHI-style batched YOLO inference

• CLIP-style embeddings

Cloud & MLOps

• Google Cloud Platform (Vertex AI)

• MLflow for model registry

• TensorRT

Technical foundation

• Image processing

• Linux proficiency

• Docker containerization

Who you are

• Autonomous — comfortable owning a topic without close supervision

• Result-oriented — you measure yourself by what ships and works

• Builder mindset — you'd rather get something running than write a perfect spec

ATTENTION UPON DROPPING YOUR APPLICATION:

Please, immediately send an email to [email protected] with subject 'Computer Vision Engineer – Your Name' sharing concrete examples of real-time computer vision work you've done in production.

We're specifically interested in:

• Real-time detection or segmentation systems you've shipped — the model architecture (RF-DETR, YOLO, or equivalent), the latency constraints, and how you met them

• Sliced/SAHI-style inference you've implemented — the use case, why tiling was needed, and how you handled the throughput trade-offs

• CLIP-style embedding work — what you used the embeddings for (classification, retrieval, attribute detection) and how it performed in production

• TensorRT optimization you've done — what you converted, the speedup you achieved, and any precision or compatibility issues you solved

• Model lifecycle setups you've built or maintained on GCP (Vertex AI) and MLflow — how models moved from training to production

• Deployments on Linux/Docker in constrained or edge environments — especially anything running on or near physical hardware