Data Scientist – ECG Signal Processing & AI Development

Please ‼️ ONLY APPLY ‼️ if you have worked with ECG (electrocardiogram) signal processing and use AI tools on a daily basis.

Remote position - EU time zone.

Job Summary:

We're hiring a Data Scientist to advance the algorithms behind our ECG analysis platform. You'll sharpen arrhythmia classification, ECG interval detection (PR, QRS, QT), and signal quality — choosing the right technique for each problem, from classical signal-processing and delineation through to machine learning.

If you bring deep, versatile biosignal-processing expertise and enjoy problems where the data doesn't arrive clean, we'd love to talk.

The ideal candidate will have experience in ECG signal processing, machine learning model development, and AI-driven diagnostics in healthcare applications.

Key Responsibilities:

• Improve and extend the arrhythmia-classification algorithms, raising accuracy and robustness on messy, real-world recordings.
• Advance ECG interval detection (PR, QRS, QT/QTc, RR) across resting ECG, selecting the right method for each problem rather than forcing one approach.
• Strengthen signal quality through denoising, baseline-wander removal, and artifact suppression.
• Build delineation and interval-measurement methods that hold up with little or no labelled data, and design sound strategies to validate them.
• Define and run algorithm evaluation against clinical ground truth — reporting classification metrics and interval-level error, and iterating to clinical-grade performance.
• Partner with the ECG technician and annotation team to specify, review, and improve labelling workflows and the quality of ground-truth data.
• Document algorithms, datasets, and experiments to support quality and regulatory requirements.

Required skills (must-have):

• Strong Python and the scientific stack (NumPy, SciPy, pandas, scikit-learn).
• Broad, hands-on command of biosignal processing for ECG — not limited to any single family. This spans wavelet transforms (DWT / SWT / CWT / wavelet packets), derivative- and filter-based delineation (e.g., Pan-Tompkins-style QRS detection), template matching, and Hilbert-transform and other time-frequency methods — plus the judgement to choose the right technique for each problem.
• Experience building ECG delineation and interval-measurement that works with little or no labelled data — classical, rule-based, or model-based methods that need no training labels, together with sound strategies for validating them.
• Practical approaches to label scarcity: semi-supervised, self-supervised, and weakly-supervised learning, transfer learning from publicly annotated datasets, and designing annotation pipelines with clinical experts to create ground truth where none exists.
• Familiarity with the limited publicly annotated ECG resources for delineation and intervals (e.g., PhysioNet QT Database, LUDB, PTB-XL) and a clear understanding of their limitations.
• Deep learning with PyTorch and/or TensorFlow/Keras for biosignals (1-D CNNs, RNN / LSTM, transformers), where labelled data supports it.
• Demonstrated experience working with ECG or other biosignal data.
• Sound evaluation discipline for imbalanced and limited clinical data — sensitivity, specificity, PPV, F1, and AUC for classification, plus interval-level error metrics (e.g., mean and SD of onset/offset deviation against a reference) — and an understanding of why these matter in a medical context.
• Version control (Git) and reproducible, well-documented experiment workflows.
• Comfortable working alongside AI assistants such as Claude as part of a modern development workflow — using them to accelerate prototyping, coding, and research while applying sound judgement to their output.

Preferred (nice-to-have):

• Domain knowledge of ECG morphology, fiducial points (P, QRS, T), and clinical intervals.
• Experience with probabilistic sequence models for delineation (e.g., hidden Markov models) and other model-based approaches that perform well with limited data.
• Familiarity with further DSP and decomposition methods (e.g., empirical mode decomposition, adaptive filtering) to draw on alongside the core toolkit.
• Experience with heart-rate-variability (HRV) analysis.
• Exposure to MLOps and deploying algorithms into production, on-device, or edge environments.
• Awareness of medical-device software standards and regulation (ISO 13485, IEC 62304, CE / MDR), including the handling of SOUP components.
• Experience collaborating with clinical annotators and clinicians.

This is a full-time role, but other options (half-time, advisory, hourly) could be reviewed if the profile is a perfect match.