A Turkish Medical AI Model Trained on 167K Real Patient-Doctor Conversations

What is it about?

This study presents the first large-scale Turkish medical language model, trained on 167,732 real patient-doctor question–answer pairs collected from a verified Turkish medical advice platform. The dataset captures authentic language used in everyday clinical communication, allowing the model to understand and respond naturally in Turkish. Using modern techniques such as Low-Rank Adaptation (LoRA) and spherical linear interpolation (Slerp) merging, the model was fine-tuned on top of open large language models (LLMs) to improve medical reasoning and linguistic accuracy. The resulting Turkish Medical LLM can assist doctors, patients, and researchers by providing reliable, context-aware responses in Turkish healthcare settings.

Featured Image

Photo by National Cancer Institute on Unsplash

Photo by National Cancer Institute on Unsplash

Why is it important?

Medical AI tools are typically developed in English, leaving low-resource languages like Turkish underserved. This work addresses that gap by creating a domain-specific model trained on high-quality, native Turkish medical data. The open dataset of 167K real interactions provides an unprecedented resource for evaluating and training medical language models in Turkish. The study also demonstrates that local, data-driven adaptation can significantly improve healthcare accessibility, data quality, and patient safety in multilingual contexts.

Perspectives