Beyond Doctor GPT: Fine-Tuning Open-Source LLMs for Real-World Clinical Data Research

Real world healthcare data is a goldmine for medical research, but nearly 80% of it is trapped in unstructured notes, making it largely inaccessible. While large language models like ChatGPT have transformed the way we interact with machines and analyze text, their use in healthcare—especially in Hebrew—faces major challenges due to privacy, regulation, computational limits and domain expertise. In this talk, I’ll share how we fine-tuned an open-source LLM on over 500,000 Hebrew medical notes, enabling secure extraction of clinical insights directly within hospital premises. You’ll gain some practical strategies for training and serving LLMs in sensitive environments, such as knowledge distillation, model quantization and Parameter Efficient Fine-tuning (PEFT), as well as best practices on how to collaborate closely with clinical experts and medical annotators to achieve better outcomes.

Speakers