Diagnocat: Техническое собеседование

Выводы и как готовиться

• The round checks both ML coding and metric judgment: tensor-shape discipline in attention and product-aware detection evaluation.
• The detection discussion is the reusable trainer piece: naive IoU aggregation ignores duplicate predictions and missed ground truth.
• Diagnocat positions the ML role as full-cycle medical CV: clinical requirements, annotation design, model training, clinical tests, regulatory validation, deploy and monitoring.