← Papers & projects

Research preprint · 2026 · Under review

RareDx

Controlled knowledge integration and graph-grounded policy optimization for rare-disease diagnosis

40.76%
27B diagnosis Hit@10 with SFT and retrieval

Advised by Prof. Tianyu Liu at Tsinghua University, I co-designed rewards for partial diagnostic matches using disease-ontology distance, biomedical similarity, and phenotype–disease links. With SFT and retrieval, the 27B model’s diagnosis Hit@10 increased from 28.53% to 40.76%.