Roberto Cruz

The authors demonstrate that fine-tuning a two-stage retrieval system using synthetic data generated by large language models can significantly improve the performance of medical semantic search for clinical coding in multiple non-English languages, often surpassing existing English-trained models.