I am building a deep-learning pipeline that can automatically identify multiple arrhythmia classes from raw ECG recordings. The part that needs the most attention right now is model training and performance evaluation, and I want this phase handled with a Convolutional Neural Network. Because I do not yet have a clean, labeled dataset, the first step will be to assemble or create one. If you have experience sourcing or annotating ECG data—perhaps from public repositories such as MIT-BIH—or if you can guide a lightweight in-house labeling process, please outline your preferred approach. Once a reliable dataset is in place, I need a robust CNN architecture implemented in Python with a mainstream framework such as TensorFlow, Keras, or PyTorch. The network should be trained, validated, and stress-tested with appropriate metrics (accuracy, F1, sensitivity, specificity, confusion matrix). I am aiming for a model that generalises well and can be reproduced easily. Deliverables • Curated and clearly documented labeled ECG dataset (or links plus preprocessing scripts) • Training code, configuration files, and a requirements.txt or environment.yml • Trained CNN weights and an inference script/notebook that accepts new ECG traces and outputs the predicted rhythm class • Evaluation report summarising results on a hold-out test set, including graphs and confusion matrix • Brief hand-off guide so I can retrain or fine-tune the model later I value clean, well-commented code, reproducible experiments, and transparent discussion of design choices. If this fits your expertise, let’s talk through the timeline and any questions you might have.