Awan AI TechTCM Benchmark: LLMs and Doctors Comparison

largest open-source TCM benchmark with 30k+ questions evaluating 15+ models against crowdsourced expert physician baselines.

Recent Work

My most recent work in Awan AI involves creating a transformer-based model that predicts Traditional Chinese Medicine herb formulas based on symptoms and patient demographics. A complex multi-label classification problem with 1464 possible formulas. I processed 60,000+ clinical cases with 2,100 features and implemented a transformer-based model. The model achieved 98.8% per-label accuracy (Hamming Loss 0.0119) and captured 96% of correct herbs in its top-15 predictions. This model is wrapped in a Gradio web interface and is being verified by doctors.