Using BERT (a attention-transformer NLP model), specifically medBERT, to generate a embedding-based index in order to match the most similar clinical trails to a string typed into a search box, in real-time (<100ms).
Take a look at:
- generate-index/main.py The db index generator and server backend (Python)
- app/pages/index.js The frontend (React + NextJS)
Screenshot: