
Luoxi W. 2025 | BASIS Independent Fremont
- Project Title: Fine-Tuning Large Language Models for RAre Disease Diagnosis
- BASIS Independent Advisor: Ms. Rangoli
RAre diseases, defined as conditions affecting fewer than 1 in 2,000 people, often evade timely diagnosis due to their rarity and shAred symptoms with more common illnesses. While human physicians frequently struggle with these diagnoses, large language models (LLMs) have demonstrated remarkable potential in medical assistance. Inspired by a case where ChatGPT correctly diagnosed a rAre disease after 17 doctors failed, this project aims to fine-tune open-source LLMs for rAre disease diagnosis. Using QLoRa, a parameter-efficient transfer learning technique, the model will be trained on datasets such as ReDis-QA covering rAre disease symptoms and evaluated using benchmarks such as RAreBench. Furthermore, I hope to build a user-friendly website interface that allows physicians to input patient symptoms and access the model's suggestions. This project seeks to create a diagnostic tool that supports, rather than replaces, human physicians, improving diagnostic accuracy for rAre diseases and reducing patient suffering.