The integration of artificial intelligence (AI) into decision-making processes has transformative potential - but its true potential emerges when AI complements human strengths and mitigates human weaknesses. To unlock this potential, our study adopts a comprehensive approach to human-AI collaboration: Building AI to complement human weaknesses and aligning human reliance with AI capabilities through nudging strategies. We trained a complementary AI model specialized in classifying ECG signals that are particularly difficult for humans, and evaluated its performance against a baseline model. To further investigate human-AI collaboration, we conducted an online study where humans classified the same ECG signals to examine how different nudging strategies affect human reliance on suggestions from the AI. Our findings demonstrate that prioritizing human-difficult data in AI training is an effective strategy in resource-constrained settings, as these cases maximize AI's complementary potential. Additionally, the study highlights the importance of task-specific nudging strategies. While AI suggestions strongly influence user behavior, their effectiveness depends on task complexity and the accuracy of the human and the AI. In our online survey with 96 participants, Intelligent Nudging, as a tailored mechanism, enhances the accuracy of the human-AI collaboration by 20\%. In conclusion, embedding complementarity into AI training and adopting Intelligent Nudging offers significant advancements in human-AI collaboration. These approaches provide actionable pathways for optimizing decision-making, particularly in high-stakes fields like medical diagnostics.