This blog post explores how biases in health-specific AI models trained on Global North data limit their effectiveness in South Asia. We found that fine-tuning on local patient records improves performance, highlighting the need for representative datasets and regional collaboration to build equitable, accurate AI tools for global healthcare.