Computer Vision Engineer (m/w/d)

Tasks

YOUR MISSION AT CERPRO

Technical drawings are the backbone of the manufacturing industry and at the same time one of the toughest nuts to crack in computer vision: dense layouts, tiny text, hundreds of symbols per sheet, scanned legacy stock alongside clean vector PDFs. This is exactly where you come in. As a Computer Vision Engineer, you build the models that turn these drawings into usable data: dimensions, tolerances, GD&T, surface specifications, welding symbols, bills of materials. You work at the heart of our produc and help decide how far our automation can go.

YOUR RESPONSIBILITIES

• Models for technical drawings: You develop and train models for detection, classification, and segmentation of the elements on technical drawings, from individual dimension entries to GD&T symbols to complete title blocks and bills of materials.

• OCR for the technical context: You build robust OCR pipelines that handle the peculiarities of technical drawings: special characters, rotated text, low resolutions, scanned originals, mixed fonts, and safety-critical dimensions where every digit counts.

• Document understanding & layout: You work on models that understand the structure of a drawing, which annotation belongs to which dimension, which view shows which surface, where the title block ends and the bill of materials begins. For this, you use modern Document-AI approaches and vision-language models.

• Data strategy & annotation: You help decide which data we need, how we collect and annotate it, and where synthetic data can meaningfully extend our training set. You work closely with our annotators and define guidelines.

• Evaluation & iteration: You define meaningful metrics, build evaluation pipelines, and ensure that we improve model quality not just subjectively but measurably. You systematically analyze failure cases and derive the next training cycles from them.

• Deployment & performance: Together with the engineering team, you bring your models into production: latency, cost, and stability are just as important to you as accuracy. Where needed, you optimize models with quantization or ONNX/TensorRT.

• Collaboration within the ML team: You work closely with the rest of the ML team on hybrid approaches in which vision models and rule-based evaluation go hand in hand.

Requirements

WHAT YOU BRING

• Several years of experience in computer vision and deep learning, ideally in a production environment.

• Strong command of Python and one of the common DL frameworks (PyTorch preferred).

• Practical experience with object detection, segmentation, and/or OCR, including training, evaluation, and deployment of your own models.

• Familiarity with modern architectures (e.g. DETR, YOLO variants, SAM, LayoutLM, Donut) and a sense of when which approach fits.

• Experience working with documents (PDFs, scans) is a clear plus, experience with technical drawings, plans, or engineering documents is an even bigger one.

• Structured, results-oriented way of working, you don't just build models, you bring them into production.

• Basic understanding of MLOps topics: versioning, reproducibility, monitoring of models in live operation.

• Business-fluent German or English skills.