Machine Learning Engineer

Work at the core of what makes Dentail AI clinically credible — the detection models that identify pathologies and other findings in dental radiographs with accuracy that meets or exceeds specialist performance. You'll own the full ML lifecycle from research and training through to production inference and continuous evaluation. Dentail's AI engines are an ensemble of several ML models interacting through interconnected logical layers, so this role rewards both depth in modeling and systems thinking. What you’ll do Design custom model architectures for dental imaging — combining and extending convolutional, transformer-based, and multi-task approaches for detection of caries, periapical pathology, bone loss, and other findings across intraoral and panoramic radiographs. Maintain and improve our training and evaluation infrastructure — data pipelines, annotation tooling, experiment tracking, and model versioning. Collaborate with Dr. Alexander Johansson (CMO) and our medical advisory network to design clinically meaningful evaluation benchmarks and translate peer-reviewed findings into concrete model improvements. Drive rigorous model evaluation — precision, recall, calibration, and per-pathology sensitivity/specificity trade-offs on real clinical data. Optimize model inference for production: latency, throughput, quantization, and hardware-efficient deployment. Contribute to research direction: what to build next, which problems are tractable, and how to measure success.

What we’re looking for MSc in a relevant field (computer vision, deep learning, or similar); PhD welcome. 4+ years of machine learning experience focused on computer vision, with at least 2 years shipping models in production. Deep proficiency with PyTorch; experience with object detection frameworks (YOLO, DETR, or similar) and inference optimization (ONNX, TensorRT, quantization) a plus. Track record of designing or significantly modifying model architectures for specific problems — not just fine-tuning off-the-shelf models. Strong understanding of evaluation methodology — not just benchmark metrics, but sensitivity/specificity trade-offs in high-stakes classification. Familiarity with medical imaging (DICOM, radiograph modalities) is an advantage.